Newsletter of the European Network of Excellence ... - Service - DFKI

Newsletter of the 

European Network 

http://www.planet-noe.org 

of Excellence 

in 

AI Planning 

Issue No.5 

ISSN 1610-0204

PLANET News 

Issue No. 5, December 2002 

Copyright (C) 2002 

PLANET, the European Network of 

Excellence in AI Planning 

Printed in Ulm, Germany 

ISSN 1610-0204

The PLANET Newsletter 3 

Welcome to PLANET NEWS! 

Welcome to Issue No 5 ! 

PLANET’s Industrial Information Days provide a forum 

where successful industrial applications of Planning 

and Scheduling technology are presented and 

promising future exploitation is discussed. Representatives 

from industry and academia use these events 

to promote the transfer of the technology and to push 

mutual exchange and co-operation for the benefit of 

both. 

The recent information day took place in Rome in 

November. You’ll find a broad selection of contributions 

from industrial speakers in the first section of 

this issue. 

Another major event was the Second PLANET International 

Summer School on AI Planning held 

in Halkidiki, Greece in September. Students of 

the school were encouraged to present their Ph.D. 

projects in a separate poster session. This session was 

very successful and showed a variety of interesting 

aspects and new approaches. A report on the school 

and several extended abstracts of the poster presentations 

are included. 

The PLANFORM project aimed at the development 

of an Open Environment for Building Planners. It 

was carried out jointly at the universities of Hudder- 

Issue No.5 

EDITORIAL 

sfield, Salford, and Durham in the United Kingdom. 

You’ll find a final report on this project at page 38. 

The second PLANET Gap-bridging Seminar was 

held in co-location with the UK PLANSIG meeting 

in November in Delft. A report on this seminar is 

provided by Tim Grant. 

Announcements, job offers and information on forthcoming 

events, in particular those to be held in conjunction 

with ICAPS in June 2003 in Trento, and an 

invitation to participate in the Supply Chain Trading 

Competition 2003 can be found in the final section. 

Wishing you a successful and Happy New Year 2003, 

Editors: 

Susanne Biundo 

Bernd Schattenberg 

Susanne Biundo Network Coordinator, Dept. of 

Artificial Intelligence, University of Ulm, Germany, 

biundo@informatik.uni-ulm.de 

Bernd Schattenberg Network Administrator, 

Dept. of Artificial Intelligence, University of Ulm, 

Germany, schatten@informatik.uni-ulm. 

de

4 The PLANET Newsletter 

Table of Contents 

Editorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 

Industrial Information Day Rome 

S. de Givry, L. Jeannin, F. Josset, J. Mattioli, 

N. Museux, and P. Savéant 

The THALES constraint programming framework 

for hard and soft real-time applications . . . . . . . . . . . 5 

M. Sanseverino 

COMPETE a Common Platform for Extended 

Project Management . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 

J.E. Spragg 

In Defence of Reactive Scheduling . . . . . . . . . . . . . 14 

B. Drabble 

Advanced Scheduling and Optimisation: Cutting the 

Costs of Manufacturing. . . . . . . . . . . . . . . . . . . . . . . .17 

2nd PLANET Summer School 

L. Compagna, G. Cortellessa, A. Farinelli, and N. Policella 

2nd International PLANET Summer School . . . . . 22 

A. Farinelli, G. Grisetti, L. Iocchi, D. Nardi, and 

R. Rosati 

Generation and Execution of Partially Correct Plans 

in Dynamic Environments . . . . . . . . . . . . . . . . . . . . . 25 

F. McNeill, A. Bundy, M. Schorlemmer 

Dynamic Ontology Refinement. . . . . . . . . . . . . . . . .27 

G. Cortellessa, N. Policella, A. Cesta, and A. Oddi 

MEXAR: Integrated AI Technologies to Support 

MARS EXPRESS Mission Planning . . . . . . . . . . . . . 29 

M. Baioletti, A. Milani, and V. Poggioni 

RDPPlan: an Extension of DPPlan for Planning with 

Interval Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 

N.E. Richardson 

Extending Operator Induction to Provide Full 

Method Sets for Hierarchical Planning Domains . 33 


O. Sapena and E. Onaindía 

The SimPlanner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 

A. Polleres 

Answer Set Planning with DLV 

. . . . . . . . . . . . . . .36 

Project Report 

T.L. McCluskey, M. Fox, and R. Aylett 

Planform: An Open Environment for Building Planners 

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 

Gap-bridging Seminar 

T. Grant 

Second PLANET Gap-Bridging Seminar . . . . . . . . 46 

Announcements and Network News 

ICAPS 2003. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51 

3rd PLANET International Summer School . . . . . . 52 

ICAPS 2003 – Doctoral Consortium . . . . . . . . . . . . 52 

ICAPS 2003 – Workshop Program. . . . . . . . . . . . . .53 

ICAPS 2003 – Tutorials . . . . . . . . . . . . . . . . . . . . . . . 55 

R. Arunachalam 

Invitation to Participate in TAC’03 - A Supply Chain 

Trading Competition . . . . . . . . . . . . . . . . . . . . . . . . . . 59 

Postdoctoral and Doctoral Positions at Australian National 

ICT Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 

Information 

The Members of PLANET . . . . . . . . . . . . . . . . . . . . 61


ARTICLE 

The THALES Constraint Programming Framework for Hard and 

Soft Real-Time Applications 

Authors: S. de Givry, L. Jeannin, F. Josset, J. Mattioli, N. Museux, and P. Savéant 

This position paper presents the constraint technology 

that has been developed at THALES since 

1997 for introducing Constraint Programming (CP) 

in THALES operational systems 1 2 . These systems 

involve combinatorial optimization problems such as 

planning and scheduling problems that can be expressed 

with finite-domain variables and constraints. 

Typical examples of THALES systems concern supervision, 

for weapon allocation, radar configuration, 

weapon deployment and aircraft sequencing. All 

these systems are subject to specific requirements 

coming from the operational constraints of embedded 

real-time systems and from the strategic context 

of Defense applications: 

1. The system involves several functions/tasks such 

as situation assessment, resource management, visualization, 

etc.; each task is periodical and the 

period can be much shorter than a second; 

2. There is a memory space limit (a few megabytes); 

3. The system has to be supported for a long time, 

typically over 20 years for Defense applications, 

including several retrofitting (functional and platform 

evolutions); 

4. The system can be reused and modified for building 

a specific system for a new client (product 

line); 

5. The development of the system must be made and 

mastered in house for reasons of confidentiality 

and market protection. 

The CP paradigm partially meets these requirements. 

A constraint model has modularity properties, i.e. 

adding/removing a constraint is easy, which enables 

an incremental development process, reducing the 

1 This work was partially funded by the EOLE project[7]. 

2 A former version of this paper appeared in [5]. 

development time and effort. CP solvers provide efficient 

algorithms through the use of global constraints. 

The declarative nature of CP enables the programmer 

to focus on the application requirements rather than 

on debugging low-level programming errors. Validated 

CP models can be reused in a product line approach. 

Unfortunately, off-the-shelf CP solvers do not provide 

any guarantee on time and space usage. The 

classical backtracking search algorithm used in CP 

does not take into account any time contract. Recently 

an effort was made to provide better search algorithms 

in CP solvers, for instance in [1, 11, 14], but 

without any explicit time contract. Our aim is to extend 

CP solver with new search features that would 

keep the same nice software engineering properties 

as for modeling. This led to develop a high-level 

language for designing search algorithms. This approach 

allows proposing a set of search primitives on 

top of the real-time finite-domain constraint solver 

Eclair c 

[13]. The resulting search algorithms are 

based on partial search methods and take into account 

the time contract explicitly. Such algorithms can take 

advantage better of platform evolutions. 

Eclair offers time and space guarantees. Deadlines 

are guaranteed by the operating system alarm and 

Eclair is able to restore a coherent state after an interruption 

in order to deliver a valid solution, or just 

a partial solution (when not all variables are instantiated). 

The memory allocation for the constraints 

is static: a global constraint model is built once and 

only parts of the model are made active and used at 

a given cyclical call. The memory consumed during 

the search is limited by using only restricted depthfirst 

search or restricted best-first search.


Partial search methods are anytime algorithms [17] 

based on tree search methods having better quality 

profiles than the classical backtracking search algorithm. 

The main idea is to apply some arbitrary limits 

on the nodes visited in the tree search 3 , depending on 

the behavior of the heuristics and on the remaining 

computation time. We distinguish four approaches: 

the iterative weakening methods (e.g. [8]), the realtime 

search methods (e.g. [10]), the iterative sampling 

methods (e.g. [6]) and the interleaving methods 

(e.g. [12]). These methods use one or several search 

schemes 4 . The practical complexity of the search 

can be increasing, self-adjusting, or stable. In [3], we 

propose the notion of parameterized search applied 

to one search scheme. The parameters of the search 

limits are given explicitly. We can tune the degree 

of incompleteness of the search by varying the values 

of the parameters. A tuning policy indicates the relevant 

values of the parameters for different time contracts. 

In [4], we integrate the parameterized search 

approach into a hybridization scheme to express partial 

search based on several search schemes. The hybridization 

scheme is a sequence or an interleaving 

of parameterized searches. The searches can cooperate 

by exchanging solutions. A time-sharing policy 

specifies how to distribute the time contract to the 

searches. 

Our constraint optimization framework is called 

ToOLS c (Templates Of On-Line Search). A search 

algorithm is expressed in ToOLS as the conjunction 

of four distinct components: 

A set of heuristics to rank every choice; 

A set of primitives to express a search scheme 

independent of any time limit; it is composed 

by predefined choice points and combinations of 

choice points as in the OPL language [9]; 

A set of primitives to express the search limits that 

depend on the current node, the current path or 

the current sub-tree; the resulting parameterized 

search algorithm controls the size of the explored 

search tree defined by one search scheme; 

A temporal strategy defined by a hybridization 

scheme, i.e. a cooperation of several parameterized 

searches, dealing with time allocation and 

selecting the tuning strategy of the parameters 

(static tuning, iterative tuning or adaptive tuning). 

A template of search defines an abstract component 

of a search algorithm that can be reused to speed up 

the development process of customized partial search 

algorithms. This framework makes it easier to try 

new combinations of search limits and new temporal 

strategies. 

Experiments on the weapon allocation problem show 

that partial search algorithms significantly improve 

the solution quality compared to a traditional approach 

[3] and also demonstrates the gain in development 

time of new customized search algorithms. 

The code is clearer and more concise when using the 

search primitives. As the main result, our CP framework 

has been integrated in an operational on-board 

hard real-time system of THALES. 

The hybridization scheme is a way to define specific 

local search methods, such as large neighborhood 

search based on a sequence of partial searches 

in different neighborhoods [15, 16]. Pure local search 

methods could also be introduced in our framework 

as a black-box used by the hybridization scheme. The 

temporal control could be enhanced by an on-line 

learning mechanism, using the fact that similar problems 

are repeatedly solved in a real-time system. [2] 

gave the base for this mechanism. 

Bibliography 

[1] Beldiceanu, N., Bourreau, E., Simonis, H., and 

Rivreau, D. Introduction de métaheuristiques 

dans CHIP. In Proc. of MIC-98., 1998. 

[2] Crawford, Lara S., Fromherz, Markus P.J., 

Guettier, Christophe, Shang, Yi. A Framework 

3 

This description of partial search is compatible with the depth-first search principle. In [14], partial search methods are based on 

the order of node exploration, which is memory consuming. 

4 

A search scheme is a procedure that describes a search tree. For example, a combination of choice points. 

http://www.planet-noe.org


for On-line Adaptive Control of Problem Solving. 

In Proc. of CP-2001 workshop on On- 

Line combinatorial problem solving and Constraint 

Programming, Paphos, Cyprus, December 

2001. 

[3] de Givry, Simon, Savéant, Pierre, Jourdan, 

Jean. Optimization combinatoire en temps 

limité: Depth first branch and bound adaptatif. 

In Proc. of JFPLC-99, pages 161-178, Lyon, 

France, 1999 (in french). 

[4] de Givry, S., Hamadi, Y., Mattioli, J., Lemaître, 

M., Verfaillie, G., Aggoun, A., Gouachi, I., 

Benoist, T., Bourreau, E., Laburthe, F., David, 

P., Loudni, S., Bourgault, S. Towards an on-line 

optimization framework. In Proc. of CP-2001 

workshop on On-Line combinatorial problem 

solving and Constraint Programming, Paphos, 

Cyprus, December 2001. http://www.lcr. 

thomson-csf.com/projects/www_eole/ 

workshop/olcp01-eole.ps 

[5] de Givry, S., Gérard, P., Jeannin, L., Mattioli, J., 

Museux, N., Savéant, P. A constraint optimization 

framework for real-time applications. In 

Proc. of AIPS 2002 workshop on On-line Planning 

and Scheduling, Toulouse, France, April 

2002. 

[6] Gomes, C., Selman, B., Kautz, H. Boosting 

Combinatorial Search Through Randomization. 

In Proc. of the 15th National Conference on Artificial 

Intelligence (AAAI-98), pages 431–437, 

Madison, WI, USA, 1998. 

[7] RNRT EOLE project. http://www.lcr. 

thomson-csf.com/projects/www_eole 

[8] Harvey, William D., Ginsberg, Matthew L. Limited 

discrepancy search. In Proc. of IJCAI-95, 

pages 607-613, Montréal, Canada, 1995. 

[9] Van Hentenryck, P. OPL: The Optimization 

Programming Language. The MIT 

Press,Cambridge,Mass., 1999. 

[10] Korf, Richard E. Real-time heuristic search. Artificial 

Intelligence, 42:189-211, 1990. 

[11] Laburthe, François. SaLSA: a language for 

search algorithms. In Proc. of CP-98, pages 

310-324, Pisa, Italy, October 26-30 1998. 

[12] Meseguer, P. Interleaved depth-first search. In 

Proc. of IJCAI-97, Nagoya, Japan, pp. 1382- 

1387, 1997. 

[13] PLATON Team. Eclair reference manual, 

Version 6.0. Technical Report Platon-01.16, 

THALES Research and Technology, Orsay, 

France, 2001. 

[14] Perron, L. Search Procedures and Parallelism 

in Constraint Programming. In Proc. of CP-99, 

pages 346-360, Alexandria, Virginia, 1999. 

[15] Pesant, G. and Gendreau, M. A Constraint Programming 

Framework for Local Search Methods. 

Journal of Heuristics, 1999. 

[16] Shaw, P. Using constraint programming and 

local search methods to solve vehicle routing 

problems. In Proc. of CP-98, pp. 417-431, Pisa, 

Italy, 1998. 

[17] Zilberstein, Shlomo. Using Anytime Algorithms 

in Intelligent Systems. AI Magazine, 17(3):73- 

83, 1996. 

Author Information 

Simon de Givry INRA / Biometrics and Artificial 

Intelligence, Chemin de Borde Rouge, BP27, 

31326 Castanet-Tolosan cedex France, degivry@ 

toulouse.inra.fr 

Laurent Jeannin, François-Xavier Josset, 

Juliette Mattioli, Nicolas Museux, Pierre 

Savéant THALES Research & Technology / 

PLATON Center, Domaine de Corbeville 91404 

Orsay cedex France, laurent.jeannin@ 

thalesgroup.com


ARTICLE 

COMPETE a Common Platform for Extended Project Management 

Abstract 

COMPETE (Common Platform for the ExTended 

Enterprise) is an ESPRIT project which started in 

1999 and ended at the end of 2001. It is the result 

of the cooperation among different European 

partners: Centro Ricerche Fiat, TXT e-solutions, 

Cap Gemini & Ernst & Young, BAE Systems and 

Magneti Marelli, University Federico II of Naples. 

Its main objective is to integrate methods and tools 

to support process/project management in an extended 

enterprise. In this environment process 

know-how (how to do things) and available competences 

play a central role to shorten product development 

time and to support rapid decisions and 

flexibility in process actuation. 

COMPETE has developed a software platform 

which integrates process modeling methods, competences 

management, project management and 

workflow through a common data model shared by 

all company departments. 

This paper , starting from the description of current 

companies needs, describes COMPETE approach 

and functionalities and gives an idea of its general 

architecture. 

Keywords: Project Management, Knowledge 

Management, Competencies, Workflow, Business 

Processes, Modeling, Simulation. 

Introduction 

The competitive arena in the Product Development 

scenario is marked today by rapidly evolving technologies, 

dynamic, sophisticated and global markets, 

requiring high and customized performances at low 

cost. This trend has put a tremendous pressure on the 

design process which must supply a stream of products 

with high value/cost ratios at a rate fast enough to 

withstand technology obsolescence and demand vagaries. 

The main goal of COMPETE is to provide IT based 


Author: M. Sanseverino 

methodologies and tools to help companies structured 

as Extended Enterprises to cope with the challenges 

of globalization, deregulation and contracting 

life cycles, combining fast decision making and 

flexibility to change. Such methodologies embrace 

both product function, market, life cycle analysis, together 

with organisational and individual competencies 

identification and evaluation. The IT tools to be 

integrated in a distributed Extended Enterprise architecture 

(called BMA Business Modelling and Actuation 

) are: a Business Process Definition toolset (with 

modelling and simulation capabilities), a Competencies 

Identification and Evaluation environment, a Human 

Resources Planner and Scheduler, a Commercial 

Project Management tool (OPENPLAN) and a Workflow 

Management tool. 

Three main areas are targeted: the product, the process, 

and the competencies of human resources (p 3 

paradigm: product, people, process). 

On the product side, the project is concerned with 

the early conceptual phase where the links between 

functional specifications and customer’s value are 

identified and exploited to achieve maximum value at 

minimum cost, ensuring minimal environmental impact 

as well. 

Figure 1: Scenario and objectives of COMPETE


On the people side, the project is concerned with 

a system for managing human resource competencies 

in the short and medium term. Given that process 

quality and competencies are correlated, an ideal 

competencies profile is associated to each activity 

and candidates for the activity are selected who better 

match this profile. To support this procedure, competencies 

are routinely evaluated, and compared with 

medium term requirements to orientate human resources 

policies. Specific methodologies have been 

implemented to find organizational competencies ans 

evaluate individuals 

On the process side, the project is concerned with the 

generic product development process (PDP) from the 

early product concept to the detailed design. To slash 

costs and development times and to improve quality, 

enterprises are resorting to concurrency, BPR reengineering, 

standardisation, process control, knowledge 

distribution, outsourcing and extended enterprising. 

Stand alone tools have appeared to support this trend. 

The project provides a platform which supports these 

new approaches, by integrating the existing tools, 

based on a sound logical model. 

Process modelling is crucial to achieve process effectiveness. 

Actual processes are assembled from 

standard templates, which consolidate the best prac- 

tices. Processes are released for execution after detailed 

simulation and planning, which balance effectiveness 

and competencies level. The unique combination 

of a Modeller, a Wf Management Tool and 

a Project Management tool caters for tight process 

control, reactiveness to unforeseen events, high quality 

and reduced processing times. The full potential 

of this layout can be obtained by proper interfacing 

with a PDM system. The systematic exploitation of 

the existing interfacing standards and use of a Web 

based architecture makes the platform ideally suited 

to support extended enterprise arrangements. 

COMPETE Functions 

Figure 2: Functional diagram of COMPETE Platform 

The main functions supported by COMPETE platform 

are: 

1. To store the operational know-how by means 

of process modelling and to support the choise 

among different project alternatives estimating 

costs and times by means of simulation. 

2. To support the automatic translation of a process 

model into operational projects and into workflow 

flow-charts.


3. To support planning according to competencies 

required by process activities and automatically 

assigning resources according to workload in a 

multiproject environment. 

4. To support progress control integrated with 

project management. 

5. To integrate different company departments, such 

as human resources department, design, product 

planning, project management, working teams, on 

a common platform and using a common data 

model. 

6. To manage in a unique environment planning and 

progress control of distributed teams in an extended 

enterprise. 

The platform integrates a project management tool 

(OPENPLAN) and commercial workflow management 

tools with new developments according to manufacturing 

companies requirements and has the following 

advantages: 

1. On the process side 

A clear structuring of operational know-how 

for porpuse of documenting, distributing and 

consolidating knowledge 

A poweful decisional support, based on cost 

and time indicators, obtained by means of simulation 

of different project alternatives 

2. On the competencies management side 

Availability of a catalog of competencies belonging 

to the organisations and to the individuals 

of distributed teams, in a common 

database and using a common glossary 

Easy evaluation of competency needs 

Support to the definition of careers and resources 

outsourcing 

Support to training plan definition 

Efficient resource allocation and conflict reduction 

Improvement of product quality by means of a 

better work share among working teams 


3. On the project management side 

A higher reactiveness to emerging problems, 

by means of a deeper and structured knowledge 

of possible alternatives and available resources. 

COMPETE Operational Flow 

The operational flow supported by COMPETE project 

starts from process modeling, transforms process into 

operational projects, identifying optimal choices in 

terms of time and costs, supports activity planning 

and resources assignment according to a matching algorithm 

which compares the need of competencies 

of process activities with the availability of competencies 

in human resources according to their workload. 

In the end COMPETE supports project executions 

integrating project teams executed steps with 

the project manager. 

Process Modelling and Simulation The first 

step is process modeling. At this point it is important 

to explain which is the difference between a process 

and a project. A process describes a set of activities 

and takes into account possible alternatives related 

to sudden events which change the pre-defined 

flow. Unsatisfactory results of certain activities or 

new information coming from the market may require 

some changes or reworking or outsourcing. A 

process model foresees these alternatives and supports 

a preventive analysis of the different solutions. 

It is possible to describe these alternatives with modeling 

formalisms, COMPETE supports IDEF, which 

can be used by a simulation tool which estimates 

times and costs of the different solutions. This decision 

support is very important to react more rapidly


to problems during project execution. A process can 

generate many projects each one representing the activity 

flow of one alternative. 

COMPETE supports then the automatic translation of 

IDEF models into Gantt diagrams used by a project 

manager tool. 

The modeler can start from scratch or use pre-defined 

process templates that are available in the platform. 

New processes can be created as combination of existing 

ones. IDEF formalism is used to describe processes, 

sub-processes, activities and links. Each activity 

includes the description of its inputs, outputs, 

required competencies and duration. By means of 

simulation it is possible to choose the best project, 

according to times, costs and competences availability, 

comparing the different gantt charts. 

Planning and Resource Scheduling 

Once the best project solution has been chosen, human 

resources will be allocated. 

COMPETE has developed a matching algorithm that, 

according to the competencies required by each activity, 

selects the proper candidates comparing their 

individual competencies with the required ones. In 

a second step the system compares required timings 

with individual workloads on a multi-project basis 

and proceeds in building working teams. A red 

traffic-light informs the project manager the missing 

of candidates for the required time interval and 

the necessity of acquisition, outsourcing or training. 

Manual adjustments are obviously possible, considering 

that human resource scheduling need to be 

much more flexible than machine scheduling. At this 

purpose an efficiency factor has been introduced to 

consider the fact that human resources with the same 

competencies can have speed and quality very different. 

Human resources assignment are then exported in a 

project management tool which will represent graphically, 

gantt and workload. 

Figure 4: IDEF modeling of process and competences


Progress Control 

Once the planning and the resource scheduling are 

completed, the working teams are ready to start working. 

COMPETE supports in this phase the aligning 

between the operative management and the progress 

control. The activity flow, firs modeled and then 

planned is automatically imported in a workflow 

manager tool which supports activity dispatching. 

The activity manager will insert in the system information 

related to activity status and progress and the 

project manager tool will be automatically updated 

according to these data. 

Competencies Management 

COMPETE stresses the importance of the relation between 

competencies management and process quality. 

A competence is a combination of knowledge, 

capacity to apply it and to make it applied. It is essential 

for a company to model its competencies and 


to relate them to human resources. It is essential in an 

organization to compare its competencies with process 

and project needs. 

COMPETE has developed a set of tools and a database 

to create a competence tree, to describe the organization, 

to relate competencies, organization and human 

resources and to evaluate human resources according 

to their competencies. It is very important for a 

company to build its competence tree in order to understand 

which competencies are available and which 

one need to be acquired in order to face technological 

evolution. 

COMPETE Architecture 

Figure 5: COMPETE Architecture 

The improved prototypes developed in COMPETE are 

currently being engineered and packaged into a product 

suite composed by two main tools: SKILLPLAN 

(Skill & Process Planner) and P-CON (Process Actuation 

Control & Progress Review)


SKILLPLAN supports the process planning and optimisation, 

by means of process modelling, specification 

of competencies’ requirements, simulation and 

what-if analysis. Process simulation caters for alternative 

flows of activities and recycling probabilities. 

The plan is then turned into operation: activities 

are assigned to resources according to required competencies, 

and scheduled. SKILLPLAN architecture 

is based on the XML integration of an IDEF-based 

modeller, which describes the process, a simulator 

aimed to disambiguate it analysing all the related 

possible project alternatives, a scheduler which uses 

an innovative algorithm to find the best resources, 

according to the required competencies and to the 

time schedule. A commercial Project Manager has 

been integrated: this is Open Plan (integrated into the 

suite) which supports multi-projects facilities. 

The main and innovative features of SKILLPLAN 

are: 1- the possibility to perform what-if analysis in 

order to find the best projects to implement a certain 

process, 2- the introduction of a competency based 

HR planning & Scheduling approach. 

P-CON aims at managing events, notifications and 

workflow (dispatching the activities among the actors 

involved) along the project lifecycle. It is tightly integrated 

through XML interfaces to SKILLPLAN in 

order to react as fast as possible to problems arising 

during project actuation. 

The formalisms used for description of processes and 

projects , such as IDEF, gantt, flowcharts have been 

completely integrated and can be imported and exported 

through XML interfaces. 

The architecture (Fig. 5) is applicable in an extended 

enterprise environment. The communication among 

the distributed tools of COMPETE platform is based 

on an asynchronous channel for XML models transmission 

and on a distributed database for competencies 

storage. 

Conclusion 

In conclusion a stream of new organizational approaches 

have been introduced in the recent years to 

streamline Product Development Process , to achieve 

better quality and trim the costs. Concurrent engineering, 

businesses process engineering, benchmarking, 

knowledge management, process standardization, 

Extended Enterprises have been added to more 

traditional automation tools such CAD, CAD/CAM 

and CAE. Stand alone tools have appeared to support 

these new practices. COMPETE brings together these 

new approaches in an integrated platform, which ensures 

as much as possible the communication among 

tools of different make. The benefits of the system, its 

capability to coordinate concurrent activities in distributed 

environment, are such as to significantly impact 

on quality, time-to-market and costs in modern 

extended enterprise. 

Bibliography 

[1] G. Morra, M. Sanseverino Un progetto Globale: 

il caso Fiat, Sistemi e Impresa , September, 

1996. 

[2] G. Zollo, A. Cannavacciuolo, G. Capaldo, A. 

Ventre, A. Volpe The Organisational Evaluation 

Process, Fuzzy Economic Review, 1, no.1 (1996) 

3-30. 

[3] M.Argilli, S.Gusmeroli, G. Secco Suardo, N. 

Matino, M. Sanseverino COMPETE : a common 

platform for the extended enterprise 2000, 

e-Business and e-Work 2000 , October, 2000. 

[4] G. Michellone, G.Zollo Competencies management 

in knowledge based firms, Internationa 

Journal of Technology management, vol 

20, 2000. 

[5] G. Zollo, L. Iandoli, F. Borrelli, A Iuliano, M. 

Sanseverino Applications of soft computing techniques 

to Product Design, International Seminar 

on Intelligent Computation in Manufacturing 

CIRP June, 2000. 


Marialuisa Sanseverino FIAT Research Center, 

marialuisa.sanseverino@crf.it


In Defence of Reactive Scheduling 

Introduction 

Reactive Scheduling is the Cinderella of scheduling 

techniques. Reactive Scheduling drudges away in the 

work place, delivering feasible solutions to highly 

complex and dynamic resource allocation problems 

but is seldom invited to conferences to show its finery 

to the princes of the scheduling and planning community. 

Their attention is distracted by yet another algorithm 

that promises optimum results on some static 

benchmark problem. Indeed, a recent PLANET sponsored 

workshop on on-line planning and scheduling 

[1] did not even include reactive scheduling as one 

of its topics of interest until the second call for papers. 

This is strange, on-line scheduling domains are 

the most sensitive to sudden environmental change. 

The reality is that in stochastic environments the only 

option open to a scheduler is to react to change. Techniques 

for anticipating change are limited and seldom 

provide the functionality needed to manage resource 

allocations that are continually being impacted by 

disruptions of one form or another. 

The Need for Reactive Schedulers 

The requirement for reactive scheduling systems is 

more widespread in the industrial world than most researchers 

from the academic world realise. For example, 

the production system used for garment manufacture 

in the UK is called a Progressive Bundle Line 

(PBL) [2, 3, 4]. It is a flow line manufacturing system 

in which work stations, comprising of machinist and 

sewing machine pairs, are arranged in a configuration 

so that the flow of work from one work station provides 

work for the next work station in the production 

sequence. The scheduling objective is to maintain a 

line balance so that the work in progress buffers between 

each work station never overflow, causing storage 

and quality problems, or empty, causing machin- 


ARTICLE 

Author: J.E. Spragg 

ists to sit idle at their machines. 

There are examples of line balancing algorithms in 

the scheduling and optimisation literature. The majority 

of these techniques view line balancing as a 

static optimisation problem and are wholly inadequate 

for tackling the management of a PBL. The fact 

of the matter is, in the real world, a well balanced 

line soon becomes unbalanced. Machines breakdown, 

machinists go absent or start working below, or 

above, standard performance, managers decide that 

priority job batches are no longer important and that 

low priority batches are important, quality controllers 

decide that rework is required on some batches, etc. 

In clothing factories that employ PBL production systems 

the most important person is the line supervisor 

who continuously monitors, analyses, and revises the 

flow of work through the work stations. How well 

she (it is usually a she) does this depends upon her 

experience and training. 

Providing reactive scheduling capabilities to a realtime 

automated scheduler involves mimicking the responsibilities 

of the human supervisor employed to 

manage a PBL. An ideal reactive scheduling framework 

employs an event driven multi agent approach 

that applies monitoring, analysis, revision, and optimisation 

tools in real-time to an executing schedule 

to maintain its feasibility and quality over time. 

Reactive scheduling techniques can be reinforced by 

building robust schedules or providing probabilistic 

models of system behaviour. While nobody doubts 

the importance of these techniques; if a machine is 

to breakdown it would be useful to have an indicator 

of which machine and when; there is still the problem 

of what the system can do about it. The response 

to a machine breakdown for example is often context 

sensitive and depends upon the current opportunities 

for resource reallocation within the current schedule. 

Robust schedules limit the impact of a disturbance 

on a schedule but again the system still has to reason


about strategies for bringing the solution back to feasibility. 

At the end of the day, the system will need 

to react by relaxing due dates or reallocating activities 

to alternative resources or bumping lower priority 

operations. 

Mixed Initiative Scheduling – and beyond 

The OZONE Project [5] at the Intelligent Coordination 

and Logistics Laboratory, Carnegie Mellon University, 

back in the mid-90s of last century, recognised 

that current scheduling systems do not effectively 

support user tasks and requirements and do 

not support the iterative, evolving process of problem 

understanding, requirements determination, conflict 

resolution and solution refinement that is inherent 

in large, multi criteria problem solving. OZONE 

tackled these issues by enabling flexible collaborative 

problem solving between user and system, supporting 

reconfiguration of system functionality to accommodate 

new environments and scheduling objectives. 

The OZONE approach to scheduling recognised 

that scheduling is an incremental process of ‘getting 

the constraints right’ in which human users always 

have the big-picture decision-making expertise and 

knowledge to contribute but are unable to effectively 

cope with the complexity of detailed solution development. 

The challenge for the new generation of real-time, 

on-line, schedulers is to encode the strategic knowledge 

that humans provide within an autonomous 

scheduling framework that can respond immediately 

and intelligently to change. When analysing a 

scheduling solution, both human and software agents, 

ask the same two questions: 

Where are the processing bottlenecks? 

Where are the scheduling opportunities? 

A mixed initiative scheduling tool supports user experimentation 

with graphical displays and statistical 

summaries. An autonomous scheduler can keep 

internal representations of analysis results and provide 

revision tools that are activated by internal state 

changes to the control architecture. An autonomous 

scheduler needs to evaluate the ‘correctness’ of its revisions 

by comparing the results against ‘objective’ 

criteria given by a multi-criteria cost function. The 

interpretation of such evaluations are often problematic 

and seldom objective. Techniques developed by 

decision theorists, such as criteria voting [6], for resolving 

conflicts in multi criteria decision making 

suggest promising alternatives to single valued measures 

of ‘optimality’ for self-governing systems. 

Reactive Problem Solvers – Schedulers 

for the new era 

The recent developments in scheduling technology 

have been partly driven by improvements in monitoring 

technology. Communication technology has 

made reactive scheduling more important by enabling 

schedulers to keep up to date with environmental 

change and allowing them to respond to disruptions 

by resolving conflicts in impacted parts of the solution. 

Gone are the days in manufacturing where 

jobs were ‘chased’ through the factory by a progress 

chaser who spent their day sweet talking departmental 

foreman and climbing over work inventories looking 

for job numbers. The advent of mechanically 

readable bar codes means that work can now be continuously 

monitored through a production process. 

Mobile telephony has had a major impact on logistic 

scheduling, allowing huge savings by allowing 

field workers to communicate with headquarters 

in real-time. The British Telecom mobile workforce 

scheduling problem [7] addressed by a.p.solve 

is highly dependent upon mobile telephony. It is a 

problem characterised by: 

A varying workload with one hour response times. 

A diverse mixture of operational procedures that 

range from those with hard start-time constraints 

to those with highly relaxable start-time constraints. 

The processing times of scheduling activities can 

range from a few minutes to several days and are 

subject to uncertainty. 

Resource scarcities in rare skills.


Scheduling activities can be complicated by dependency 

relationships between activities. 

The environment in which mobile workers operate 

is subject to uncertainty and change. Calculating 

travel times is problematic and subject to traffic 

conditions. 

British Telecom has a workforce of 20,000 technicians 

nation-wide with several hundred thousand 

tasks to be scheduled and executed every day. 

In a resource constrained environment, with a small 

time budget to resolve conflicts and improve solution 

quality, the system is forced to seek out spare capacity 

by searching a space of reallocation moves to find 

alternative resources and/or start times for activities 

in conflict. The most promising kinds of algorithms 

for this kind of constraint based search are those like 

Ginsberg’s and McAllester’s partial-order backtracking 

[8] that combine aspects of both systematic and 

non-systematic search. However, in practical environments 

partial-order backtracking needs to be supported 

by texture analysis that indicates the aggregate 

demand for alternative resources by providing measures 

of contention and reliance [9]. 

Given the industrial need for reactive scheduling systems, 

why is the scheduling community obsessed 

with static optimisation problems? In practice, 

scheduling problems require a ‘solution’ to be more 

than the mere implementation of an algorithm for 

solving a particular constraint satisfaction, or constrained 

optimisation problem. Constructing schedules, 

in practical environments is an extended, iterated 

process that typically involves resolving conflicts 

between competing schedule users and scheduling 

tools. In most applied domains, schedules need 

to be maintained over time through reactive revision 

that refine an initial, or current, solution, by adapting 

it to changing environmental conditions and user 

preferences. 

Bibliography 

[1] AIPS 2002 Workshop on On-line Planning and 

Scheduling, Toulouse, France, April 24th, 2002. 


[2] Fozzard, G., Spragg, J., and Tyler, D. Simulation 

of Flow Lines in Clothing Manufacture, Part 

1: model construction, Int.J. of Clothing Science 

and Technology, Vol 8, No. 4, pp 17-27, 1996. 

[3] Fozzard, G., Spragg, J., and Tyler, D. Simulation 

of Flow Lines in Clothing Manufacture, Part 2: 

credibility issues and experimentation, Int.J. of 

Clothing Science and Technology, Vol 8, No. 5, 

pp 42-50, 1996. 

[4] Spragg, J.E., Fozzard, G. and Tyler, D.J. FLEAS: 

a flowline environment for automated supervision, 

The Int.J. of Manufacturing Technology 

Management, Vol. 10, Number 6, pp 322-327, 

1999. 

[5] Smith, S., Lassila, O., and Becker, M. Configurable, 

Mixed Initiative Systems for Planning 

and Scheduling, Advanced Planning Technology, 

Technological Achievements of the ARPA/Rome 

Laboratory Planning Initiative, Edited Austin 

Tate, AAAI Press, pp 235-241, 1996. 

[6] Croce, F.E., Tsoukias, A., and Moraitis, P. Why 

is it Difficult to Make Decisions Under Multiple 

Criteria, AIPS-02, PLANET Sponsored Workshop, 

Planning and Scheduling with Multiple 

Criteria, pp 41-45, 2002. 

[7] Lesaint, D., Azarmi, N., Laithwaite, R., and 

Walker, P. Engineering Dynamic Scheduler for 

Work Manager, BT Technology Journal, Vol. 16. 

No. 3 July, pp 16-29, 1996. 

[8] Ginsberg, M.L., and McAllester, D.A. GSAT and 

Dynamic Backtracking, Knowledge Representation 

and Reasoning Conference, 1994. 

[9] Beck, J.C., Davenport, A.J., Davis, E.D., and 

Fox, S. The ODO Project: Towards a Unified 

Basis for Constraint-Directed Scheduling, J. of 

Scheduling, 1, 89-125 (1998). 


John E. Spragg a.p.solve, (www.apsolve. 

com), Sirius House, Adastral Park, Martlesham, Ipswich, 

Suffolk, IP5 3RE, United Kingdom, john. 

spragg@bt.com


ARTICLE 

Advanced Scheduling and Optimisation: Cutting the Costs of 

Manufacturing 

Abstract 

The aim of this article is to describe several intelligent 

scheduling techniques that have been applied 

to problems in manufacturing and assembly. 

The techniques have been evaluated in domains 

including aircraft wing manufacturing, fiber optic 

cable manufacturing, submarine construction and 

CD manufacturing. The article will describe two 

particular scheduling techniques: schedule packing 

and squeaky wheel optimization. The techniques 

have resulted in a number of major improvements 

including a reduction in make-span of up to 50%, 

an improvement in throughput of 40%, a reduction 

in costs of 20% and the ability to tackle problems 

up to 20 times larger. The article provides 

an overview of these techniques and describes two 

case studies from Boeing and Electric Boat. The 

article concludes with a description of the impact 

these techniques have had in each of these organizations 

and provides pointers to potential future 

improvements. 

Introduction 

Scheduling is the problem of assigning a set of tasks 

to a set of resources subject to a set of constraints. 

Examples of scheduling constraints include deadlines 

(e.g., job i must be completed by time t), resource capacities 

(e.g., there are only four drills), precedence 

constraints on the order of tasks (e.g., a piece must 

be sanded before it is painted), and priorities on tasks 

(e.g., finish job j as soon as possible while meeting 

the other deadlines). In addition the task assignments 

must also meet a number of optimization criteria, e.g. 

minimize makespan, minimize set up times, maximize 

work in progress. Examples of scheduling domains 

include classical job-shop scheduling, manufacturing 

scheduling, and transportation scheduling. 

This article describes two generic scheduling tech- 

Author: B. Drabble 

niques that were developed and applied to problems 

in aircraft manufacturing and submarine assembly. In 

each case the results obtained are currently the best 

in the world. These techniques are now being augmented 

with additional functionality to tackle problems 

involving ship repair/overhaul and mission planning 

for the USAF. The techniques can be divided 

into two main areas: 

The use of non-systematic techniques such as 

“squeaky wheel” optimization [4] (SWO) and 

schedule packing (also known as doubleback optimization) 

[2] to solve problems that arise in manufacturing 

scheduling. 

The use of combined systematic and nonsystematic 

techniques, such as limited discrepancy 

search (LDS) [3] with schedule packing, and 

squeaky wheel with operations research methods. 

Scheduling Technologies 

This section provides an overview of the scheduling 

technologies that have been deployed in a number of 

real applications. 

Limited Discrepancy Search (LDS) and 

Heuristics 

In scheduling problems of any size it is unlikely that 

always using a merely good heuristic will get you really 

close to an optimal schedule. A merely good 

heuristic will be incorrect some of the time. As 

the complexity of scheduling problems increases, the 

number of decisions guided by the heuristic also increases. 

The more decisions made, the more likely it 

is that some of them are going to be incorrect. How 

does LDS help address this problem? LDS is a systematic 

method for disregarding the recommendation of


a heuristic a limited number of times (thus the name 

LDS) when generating a schedule. With LDS, schedules 

are generated repeatedly, each of them following 

the heuristic for all decisions except one. The decision 

at which the heuristic is ignored is different in 

each schedule. If the heuristic leads to only one incorrect 

decision, then using LDS1 (the fastest form 

of LDS) will lead to a perfect schedule. Even if the 

heuristic leads to more than one incorrect decision 

(which is usually the case) then LDS1 will likely lead 

to a better solution then always following the heuristic. 

Schedule Packing 

Schedule Packing, also known as doubleback optimization 

involves “sloshing” a candidate schedule, 

repeatedly, right and left within a scheduling window. 

This has a remarkable impact on the length of most 

schedules. The Doubleback process is analogous to 

filling a box with blocks and then shaking the box. 

Shaking the box will almost always result in a denser 

packing of blocks. Likewise, in schedules, Doubleback 

almost always results in a denser packing of 

tasks in a schedule. The denser packing allows for 

tasks with few preceding activities to find appropriate 

holes in the schedule in which to be placed. The 

schedule packing algorithm is appropriate for problems 

with large numbers of precedences, e.g. assembly 

tasks, manufacturing, overhaul, etc. 

Squeaky Wheel Optimization 

The insight behind SWO is that in any real world 

problem it is impossible to capture all associated constraints, 

e.g. context information. SWO uses a priority 

queue to determine the order in which tasks should be 

released to a greedy scheduling algorithm. The priority 

queue is determined by how difficult the task is to 

deal with that is, i.e. higher the task is in the queue 

the harder it is to find a good resource assignment. 

On each iteration of the algorithm, SWO quickly creates 

a schedule and then examines it to identify the 

parts that were handled badly, for example, the task 


was completed too late or assigned to an unsuitable 

agent. Any task that “squeaks” is promoted up the 

priority queue, with the distance it is promoted determined 

by the extent of the problem. The new priority 

queue is then used to generate another schedule 

that is analyzed for problems. This process continues 

until no significant improvement in the schedule 

is noted over several iterations or a predefined limit is 

reached i.e. cycle count or elapsed time. SWO is extremely 

fast with each cycle of generate, analyze, and 

re-prioritize taking less than a second, even for large 

problems, e.g. 2500 tasks and 200 resources over a 

five day period. 

Application Domains 

This section describes two example domains that 

have been tackled using either schedule packing or 

squeaky wheel optimization. In each example a description 

of the domain will provided together with 

a description of the impact the technology has made. 

In addition to the domains described here the techniques 

have also been applied to submarine assembly, 

aircraft mission scheduling [5] and CD manufacturing. 

Full details can be found via pointer 

www.otsys.com/scheduling.html 

Aircraft Assembly 

The original aircraft assembly problems tackled by 

schedule packing were provided via a research group 

at McDonnell Douglas. These were real scheduling 

problems and were made available via the net to encourage 

academic researchers to demonstrate the applicability 

of their techniques. These particular problems 

are relevant to large scale assembly and are instances 

of problems known as resource constrained 

project scheduling (RCPS). CIRL developed a scheduler 

that produces the best known results on this problem. 

CIRL has also converted several other benchmark 

problems to the same format and solved them 

successfully. The basic aircraft assembly problem 

has the following features:


Zone resources : A zone is an area of the aircraft 

in which work can be done. Zone resources specify 

the maximum number of people that can work 

in that zone at the same time. 

Labor resources : These specify how many laborers 

are available with a particular skill set. 

Shifts : The availability of labor resources varies 

over time, with more being available during the 

day. 

Tasks : Tasks have a specified duration and set of 

zone and labor resources that are needed to perform 

the task. 

Precedences : These specify which tasks must be 

completed before other tasks can begin. 

The program uses limited discrepancy search (LDS) 

and schedule packing (also known as doubleback optimization) 

to generate solutions. LDS and schedule 

packing can be used in isolation or in combination 

with each other with the best results produced 

using LDS with schedule packing or schedule packing 

on its own. When used in combination, multiple 

seed schedules generated with LDS are fed to schedule 

packing. 

In figure 1 the top part indicates resource usage over 

time. There is one horizontal bar for each of the 17 

Figure 1: Example of aircraft assembly 

resource types. The darker areas indicates a resource 

is fully utilized, and the lighter indicates it is unused. 

The boxes in the bottom part of the figures represent 

the tasks. The width of a box represents the duration 

of a task, and the height is an indicator of how 

many resources a task requires. The PERT 1 schedule 

for this problem ends after 37 days, 2 hours and 

58 minutes, and is a loose lower bound on the minimum 

length any potential solution. The current best 

schedule produced by McDonnell Douglas (now the 

Boeing Company) is just over 42 days and the best 

schedule pack schedule is just over 39 days. Each 

day of production removed from the schedule saves 

the company approximately $600,000 and thus the 

schedule is able to save approximately $3.2 million 

per sub-assembly. On Time Systems 2 has extended 

this scheduler to handle more general aircraft manufacturing 

problems. These new constraints and features 

include: 

multiple sub-assemblies could be introduced into 

the line at fixed rates, e.g. “every five days” or at 

arbitrary points, e.g. “40 wings over the next 10 

days”. 

Once a wing assembly was assigned to a bay it 

must return to the same bay for further processing 

steps. 

1 The PERT schedule is generated by starting tasks as early as possible and ignoring all resource conflicts 

2 On Time Systems is a technology startup company that was developed to commercialize the optimization technology being de- 

veloped at CIRL and from other groups around the world.


The assembly lines have a number of robots that 

must be taken down for routine maintenance. 

Submarine Assembly 

Existing scheduling systems in use at shipyards today, 

such as ARTEMIS, SAP, or PRIMIVERA, rely 

on “makespan minimization” techniques to develop 

schedules. Specifically, they try to schedule tasks 

as early as possible, subject to constraints and resource 

availability profiles. Manual intervention is 

typically required to deal with overloaded resources 

and other difficulties, and the process of scheduling 

the construction of a single ship can take months. 

This approach relies on the conventional wisdom that 

it can never be wrong to get work done early, and 

the assumption that a short schedule is likely to be 

efficient, since otherwise the inefficiently utilized resources 

could be loaded up to get more done earlier. 

OTS) research has shown that this conventional wisdom 

is, in fact, misleading. In mass production environments, 

makespan minimization often is a useful 

approach, since other jobs can help smooth out 

the resource loading artifacts the process induces. In 

shipyards, where it is not uncommon to have one, or 

at most a few, projects in process at any one time, 

makespan turns out to be a poor stand-in for the shipyards’ 

more complex goals. 

OTS has developed a radically new approach to ship 

construction scheduling, one that addresses shipbuilding’s 

unique needs directly. The resulting system, 

ARGOS, is capable of scheduling multiple years’ 

production across a whole yard in hours, instead of 

months, without need for human intervention. The 

resulting schedules typically exhibit a 10-20% reduction 

in construction labor costs when compared with 

those in use in the yards today. Conversely, in situations 

where throughput is limited by the available 

manpower pool, ARGOS makes it possible to progress 

10 to 20% more work through the yard. All of these 

savings are achieved without changing the fundamental 

production process or shipyard facility in any way. 

Table 1 shows the expected savings (over the cur- 

3 This is the schedule the yard is currently building to. 


rent schedule) for a single hull and Table 2 shows 

the expected savings for the entire yard over the next 

5 years. Our estimates are that, if ARGOS were applied 

to all new Navy construction, annual savings 

could be expected to be between $200M and $500M. 

Numbers for re-fit and repair are more difficult to obtain, 

but the percentage savings (10-20%) appear to 

be comparable. 

iteration Time Savings 

1 2 min 8.4% $13.0M 

7 10 min 11.4% $17.7M 

20 34 min 11.8% $18.2M 

Ultimate ˜24 hrs 15.5% $24.0M 

Table 1: Expected Savings for a Single Hull 

iteration Time Savings 

1 24 min 7.8% $49M 

7 1 hour 10.2% $65M 

20 4 hours 10.7% $68M 

Ultimate 4 days 11.5% $73.0M 

Table 2: Expected Savings Over the Entire Yard 

Figure 2 provides a comparison of the resource utilization 

and manpower curves for the current schedule 

3 (top graph) and the one developed by ARGOS 

(bottom graph) . The black line represents the amount 

of manpower required on a particular day. The red 

and blue lines represent the actual manpower available 

in the shipyard schedule and ARGOS schedule 

respectively. Deviation from the black line results 

in additional costs due to overtime, under-time or 

increasing/reducing the total work-force. The AR- 

GOS schedule has a much smoother profile requiring 

fewer changes in manpower levels and provides 

the ability to recover much more easily from unexpected 

changes in project deadlines. Figure 3 provides 

a comparison between the manpower needed in 

the current schedule (red line) and the manpower required 

by the ARGOS schedule for a single hull. This 

shows a smooth resource ramp “up and down” for 

the ARGOS schedule and far less perturbation in the 

resource levels in the yard.


Figure 2: Manpower and Resource Utilization 

Comparison 

Figure 3: Single Hull Resource Utilization 

Comparison 

Summary 

Search based scheduling technologies have matured 

to the point that they are now capable of solving 

large real world problems and provide users with high 

quality solutions. Table 3 provides a summary of the 

improvement in problem size and complexity that can 

be handled by current techniques. In addition to being 

able to solve complex real world problems it is 

also interesting to note that the technology transfer 

path for these methods is short, which makes them 

easily accessible to industrial users. 

Tasks Resources Type Feasible 

1993 64 6 Job Shop No 

1996 ˜570 17 RCPS Barely 

1999 1000s Dozens RCPS Yes 

2001 10,000s Hundreds RCPS Yes 

2002 millions Hundreds RCPS Yes 

Table 3: Improvement in Problem Size and 

Complexity 

Bibliography 

[1] Aarup, M. and Arentoft, M.M. and Parrod, Y. and 

Stokes, I. and Vadon, H. and Stader, J., Optimum- 

AIV: A Knowledge-Based Planning and Scheduling 

System for Spacecraft AIV, Intelligent Scheduling, 

(eds. Zweben, M. and Fox, M.S) Morgan Kaufmann, 

Inc, 1994, pp451-469. 

[2] Crawford, J., An Approach to Resource Constrained 

Project Scheduling, in proceedings of the Artificial 

Intelligence and Manufacturing Research Planning 

Workshop, June 1996, AAAI SIGMAN. 

[3] Harvey, W. and Ginsberg, M., Limited Discrepancy 

Search, in the proceedings of Fourteenth International 

Joint Conference on Artificial Intelligence 

(IJCAI-95), July 1995, IJCAI Inc. 

[4] D. E. Joslin and D. P. Clements, Squeakywheel Optimization, 

in the proceedings of the Fifteenth National 

Conference on Artificial Intelligence, July, 

1998, AAAI Press, Menlo Park, CA. 

[5] Drabble, B., Task Decomposition Support to Reactive 

Scheduling, in the proceedings of the 5th European 

Conference on Planning (ECP-99), Durham, 

September, 1999, Springer Verlag Press, New York, 

NY. 


Brian Drabble On Time Systems, Inc, 1850 Millrace 

Drive, Suite 1, Eugene, OR 97403-1269, USA, 

drabble@otsys.com


2nd International PLANET Summer School 

After the great success of the first edition (Cyprus, 

September 2000), the 2nd International Summer 

School on AI Planning was held in Halkidiki, Greece 

on September 16-22, 2002. 

This event was one of a number of activities organized 

by PLANET (the European Network of Excellence 

in AI planning), whose main aim is to promote 

knowledge exchange among students and researchers 

who are new to the field with a view to foster interaction 

and international collaborations. More than 

fifty PhD students and researchers from academia and 

industry attended the School, which offered courses 

held by top-level, internationally-renown speakers. 

The courses were divided into eight distinct parts, 

which covered most of the current “hot” topics in AI 

planning: 

Planning under Uncertainty with Markov 

Decision Processes 

Craig Boutilier 

Markov Decision Processes (MDPs) are a widely 

used computational method for solving sequential decision 

problems involving uncertainty. This course 

provided a brief introduction to MDPs, and focused 

on the methods of representation and solution that are 

strictly related to AI planning. A greater emphasis 

was put on approaches that reduce the computational 

effort for solving MDPs through the adoption of techniques 

developed in the AI community. 

Plan-based Control of Autonomous Robots 

Michael Beetz 

Given the increasing interest in autonomous robotic 

applications and in improving the performances 

of current systems (MARTHA, XAVIER, RHINO, 

MINERVA, REMOTE AGENT), this course aimed 

to provide the attendees with a broad overview of 


REPORT 

Author: L. Compagna, G. Cortellessa, A. Farinelli, and N. Policella 

the issues involved in the plan-based control of autonomous 

robots. Michael presented the computational 

principles and basic software architectures that 

enable robots to perform complex and diverse task in 

dynamic environments. The need for an integrated 

approach among plan representation, reasoning, execution 

and learning was strongly underlined. 

Planning history and Overview 

Susanne Biundo 

These lectures offered a comprehensive introduction 

to the field of AI Planning. They reviewed existing 

planning methods (classical, heuristic search, hierarchical 

and hybrid, and logic-based), describing in 

detail their domain representations and introducing 

present and future application areas. In addition, Susanne 

also presented developments towards a systematic 

combination and integration of different planning 

methods, as well as the integration and use of techniques 

from related fields of research. The course 

was concluded by a useful historical overview of the 

AI Planning field. 

Scheduling and Planning 

Brian Drabble 

Brian presented an overview of recent developments 

in intelligent scheduling and optimization and the 

ways in which these systems and algorithms can be 

integrated with planners to develop a comprehensive 

approach to the planning and scheduling problem. 

Traditionally, scheduling and planning were viewed 

as separate research areas. However, this is a simplistic 

view, as many decisions in planning have a direct 

impact on scheduling, and viceversa. The overview 

provided details of several scheduling techniques and 

described many of their properties.


Planning and resources 

Philippe Laborie 

Philippe’s lectures began with the introduction of the 

concept of resource – any substance or (set of) object(s) 

whose cost or available quantity induce some 

constraint on the operations that use it – and by showing 

application domains for planning with resources. 

The course then proceeded with a review of the state 

of the art of planning with resources, and presented 

basic tools and planning techniques. Finally, Philippe 

also described in detail one of the most promising approach 

for dealing with resources in planning, which 

relies on the application of constraint-based techniques 

in partial-order or hierarchical task network 

(HTN) planning. 

Planners Performance Evaluation 

Maria Fox 

The goal of this course was to analyze and discuss 

methods for the evaluation of planning systems. 

Planners can be evaluated in a number of differ- 

ent ways: by analysis of their formal properties, by 

empirical comparison with other similar systems, in 

terms of the time/quality trade-offs they make, and 

so on. This course outlined the stages of a scientific 

approach to evaluation of a data set. Moreover the 

course introduced some statistical tests that can be 

used to determine the significance of a feature of a 

data set. 

Planning and Execution 

Martha Pollack 

These lectures dealt with the interesting problem of 

planning and execution applied mainly to Simple 

Temporal Problems. The speaker highlighted the fact 

that classical planning makes strong simplifying assumptions 

on the world model. In particular, in real 

environments the beliefs and goals at the base of 

the original planning problem are subject to changes, 

and hence the solution plan might loose consistency. 

Martha provided an helpful introduction to this topic, 

dwelling on the following techniques: (a) interleaving 

planning and execution; (b) monitoring plan ex-


ecution and inferring execution status; (c) recovering 

from failure by replanning; (d) maintaining the 

temporal consistency of plans during execution (plan 

dispatch); (e) managing commitments and updating 

plans in response to changes in the environment. 

Planning and the Web 

Craig Knoblock 

This course provided an interesting overview of the 

techniques involved in the context of gathering and 

integrating information from the web. These topics 

include query planning for information gathering, interactive 

planning using constraint propagation, and 

efficient execution of information gathering plans. 

Craig pointed out how information gathering from 

the web can be tackled as a planning problem, as it 

can be viewed as a process aimed at formulating a 

scheme or program for the accomplishment of some 

goal. 

Poster Session 

The School also included a poster session, during 

which attendees were able to present the recent developments 

of their work. This allowed further interaction 

between students, researchers, lecturers and organizers 

and enabled the authors (more than fifteen) 

to receive useful feedback and suggestions. 

Social Activities 

The local organizers put together a varied programme 

of social and cultural activities. These activities comprised 

an exciting excursion to the caves of Petralonia, 

a visit to the historical city of Thessaloniki, and 

a short trip to Nea Skioni, a traditional fishing village. 

This allowed the participants to get in touch 

with the ancient history, local cuisine, origins, music 

and dances of this fascinating country. 

Conclusions 

The School was a great success, providing not only 

an invaluable and up-to-date review of the recent 


developments and techniques in AI Planning and 

Scheduling, but also a friendly and informal environment 

in which interaction was facilitated and 

research collaborations fostered. Many challenging 

questions were raised during the lectures, stimulating 

discussion and promoting knowledge exchange 

amongst all of the participants. Course 

materials, long abstract of the posters and lots 

of funny photos can be found at the web page 

of the school http://cswww.essex.ac.uk/ 

PLANET/summer-school-02/. 

Acknowledgements 

The realization of this event – sponsored by PLANET 

– required the coordinated effort of several people, 

including Susanne Biundo (University of Ulm, Germany), 

Enrico Giunchiglia (University of Genoa, 

Italy), Sam Steel (University of Essex, UK), Ioannis 

Refanidis (University of Macedonia, Greece), and 

Ioannis Vlahavas (Aristotele University of Thessaloniki, 

Greece). We would also like to thank Max 

Garagnani for his help in the production of this document. 


Luca Compagna DIST - University of Genoa, 

Italy, compa@dist.unige.it 

Gabriella Cortellessa ISTC-CNR [PST], Planning 

and Scheduling Team, Institute for Cognitive 

Science and Technology, National Research Council 

of Italy, Rome, Italy, corte@ip.rm.cnr.it 

Alessandro Farinelli DIS, Dipartimento di Informatica 

e Sistemistica, Universita’ degli Studi di 

Roma “La Sapienza”, Rome, Italy, farinelli@ 

dis.uniroma1.it 

Nicola Policella DIS, Dipartimento di Informatica 

e Sistemistica, Universita’ degli Studi di Roma 

“La Sapienza”, Rome, Italy, policella@dis. 

uniroma1.it


ARTICLE 

Generation and Execution of Partially Correct Plans in Dynamic 

Environments 

Introduction 

In this work we present the recent developments of 

the approach to the design of Cognitive Robots (i.e. 

robots whose actions are driven by a formally developed 

theory of action), that are capable of performing 

tasks in a coordinated way. The logic of actions that 

we adopt is an epistemic dynamic logic, where it is 

possible to derive acyclic branching plans (branches 

corresponding to sensing actions), including primitive 

parallel actions. 

In the present work, we consider an extended notion 

of plan by admitting a simple class of cycles that arise 

from the attempt to recover from the failure states 

originated by sensing actions. The proposed extension 

allows us to address the problem of generating 

plans that handle a form of synchronization based on 

the recognition of specific situations through sensing 

actions, including forms of coordination required in 

a multi-robot scenario. 

System Architecture 

In this section we recall the layered hybrid architecture 

used for our cognitive mobile robots (see 

also [4]) displayed in Fig. 1, that has been implemented 

on several different kinds of robotic platforms, 

namely Sony AIBOs, Pioneer, and homemade 

wheeled robots. 

The deliberative level is formed by three main components: 

The Plan Execution Module that is executed 

on-line during the accomplishment of the robot’s task 

and is responsible for executing a plan by coordinating 

the primitive actions of a single robot. The Coordination 

Module that is responsible for assigning 

tasks to the robots in the team according to the current 

situation. The Plan Generation Module, that is executed 

off-line before the beginning of the robot’s mis- 

Authors: A. Farinelli, G. Grisetti, L. Iocchi, D. Nardi, and R. Rosati 

sion, and generates a set of plans to deal with some 

specific situations. 

Off-line 

Deliberative Level 

On-line 

Deliberative Level 

Operative Level 

Sensors 

KB 

Perception 

Plan 

Generation 

High-level 

Conditions 

Coordination 

Module 

World 

Model 

Plan 

Library 

Plan 

Execution 

Primitive 

Actions 

Figure 1: Layered architecture for our robots 

Plan Representation and Generation 

Actuators 

In order to address the problem of synchronize different 

robotic platforms using sensing action we used 

a particular notion of plan that we called Partially 

Strong Plan. This notion of plan is equivalent to the 

definition of strong cyclic plans given in [3]. A Partially 

Strong Plan is a plan that if terminate leads to 

a goal state. Namely, a partially strong plan is a plan 

that is not guaranteed to terminate: termination actually 

depends on the outcome of the sensing actions 

in the plan. However, if such a plan terminates, then 

it always leads to the goal. The generation of plans 

is based on the use of a modal non monotonic de- 

scription logic ¢¡¤£¦¥¨§�© 

[2]. As illustrated in [5], 

the set of models of an �¡¤£¦¥ §�© knowledge base � 

formalizing a dynamic system can be represented by 

means of a unique transition graph, called first-order 

extension (FOE) of � , which represents all the possible 

evolutions of the dynamic system. For instance, 

Fig. 2 displays some examples of portions of FOEs.


senseA (T) 

X1 

X0 

senseA (T) senseA (F) 

A 

senseA (T) 

senseA (F) 

X2 

B 

senseA (F) 

senseA (T) 

senseA (T) 

X1 

GOAL 

X0 X0 

senseA (T) 

senseA (F) 

senseA (F) 

X2 

senseA (F) 

senseA (T) 

a) b) 

c) 

The plan generation module selects a portion of the 

FOE of the KB containing only those actions that are 

necessary to achieve a goal starting from a given initial 

situation. In fact, conditional plans can in principle 

be generated in two steps (see in [5] for details). 

First, the FOE of the knowledge base is generated; 

this FOE can be seen as an action graph representing 

all possible plans starting from the initial state. Then, 

such a graph is visited, building a term (the cyclic 

conditional plan) representing a graph in which: (i) 

sensing actions generate branches; (ii) each branch 

leads either to a state satisfying the goal or to a previous 

state of the plan. However, the current implementation 

does not build the entire FOE before searching 

for the plan, but it builds the FOE starting from the 

initial state with a breadth-first technique until a goal 

state is reached. In case of sensing actions all the 

branches are required to reach a goal state. In this 

way it is possible to extract a minimal plan (with respect 

to the number of actions to be executed). 

Implementation 

We provided an implementation of our approach using 

a simulator. The situation we experimented as 

an example for explaining our plan generation and 

execution mechanism, is the dynamic exchange of 


Figure 2: Plan structure for cyclic sensing actions 

GOAL 

senseA (F) 

the role of goal keeper in the Sony Legged League 

and the application of the two-defender rule. The 

situation presented is a typical situation in the Sony 

Legged League matches in which the goal keeper 

(robot number 1) is moving away from its own goal 

and is approaching the ball to push it away, while another 

robot (robot number 2) is far away from the ball 

and it cannot help the goal keeper immediately. In 

this situation, it is more convenient for the team that 

robot 1 takes the role of attacker pushing the ball towards 

the opposite goal, while robot 2 goes back to 

defend its own goal acting as a goal keeper. However, 

in performing this role exchange the two robots must 

comply with the two defenders rule, and thus robot 

2 can enter the goal area only after robot 1 has left 

it. The problem of complying with the two-defender 

rule is solved by generating a plan in which one robot, 

before entering the goal area, must check that it is free 

(i.e. the other robot has left the area). This is achieved 

by adding in the knowledge base of the robot the 

specification of a sensing action SenseFreeArea that 

is used for verifying if the goal area is not occupied 

by any robot of the team. Even though the simulation 

cannot provide a precise characterization of all the 

aspects that influence the performance of the robot in 

the real environment, it can provide useful feedback


to the design of the coordination and plan execution 

modules for actual robots. Through this simulator we 

have verified the intended behaviour of the robots in 

each of the roles in different scenarios. 

Conclusions 

As compared with previous work in Cognitive 

Robotics this is a novel attempt to generate plans that 

include cycles in a partially known environment. As 

compared with the work on classical planning there 

is a close relationship with the work in [1], where 

only conditional plans (tree-structured) are generated. 

We are currently addressing the application of 

model checking techniques, as done in [1], also in our 

setting. Moreover, we are extending our analysis to 

generate plans with cycles that are more general than 

the ones presented in this paper. 

Bibliography 

[1] P. Bertoli, A. Cimatti, M. Roveri, and P. Traverso. 

Planning in nondeterministic domains under partial 

observability via symbolic model checking. 

In Proc. of the 17th Int. Joint Conf. on Artificial 

Intelligence (IJCAI 2001), 2001. 

Dynamic Ontology Refinement 

Introduction 

Creating a plan that is guaranteed to be executable in 

a certain domain depends not only on forming plans 

correctly but also on having a perfect understanding 

of that domain. Unfortunately, developing this understanding 

and representing it fully is possible only 

in small, static domains. In more complex environments, 

plans may be based on incomplete or incorrect 

information and hence may be unexecutable. Interaction 

with the environment through attempted execution 

of the plan leads to an enriched and fuller un- 

[2] Francesco M. Donini, Daniele Nardi, and Riccardo 

Rosati. Description logics of minimal 

knowledge and negation as failure. ACM Trans. 

on Computational Logic, 3(2):1–49, 2002. 

[3] Fausto Giunchiglia and Paolo Traverso. Planning 

as model checking. In Proc. of the 5th Eur. Conf. 

on Planning (ECP’99), 1999. 

[4] Luca Iocchi. Design and Development of Cognitive 

Robots. PhD thesis, Univ. ”La Sapienza”, 

Roma, Italy, 1999. 

[5] Luca Iocchi, Daniele Nardi, and Riccardo Rosati. 

Planning with sensing, concurrency, and exogenous 

events: logical framework and implementation. 

In Proceedings of the Seventh International 

Conference on Principles of Knowledge Representation 

and Reasoning (KR’2000), pages 678– 

689, 2000. 


Alessandro Farinelli, Giorgio Grisetti, Luca 

Iocchi, Daniele Nardi, and Riccardo Rosati 

Dipartimento di Informatica e Sistemistica, Università 

“La Sapienza”, Roma, Italy 

ARTICLE 

Authors: F. McNeill, A. Bundy, and M. Schorlemmer 

derstanding of the environment but also leads to plan 

failure. This failure could be due to part of the theory 

itself, such a missing axiom in a rule, or due to the 

underlying ontology, such as a predicate with an incorrect 

arity. We therefore propose to devise a system 

that can dynamically incorporate this new knowledge 

into the theory as the plan is being executed. 

Our system will be based around a central planimplementation 

agent, who will control all the other 

components of the system. This agent will firstly 

send the theory, together with the goal, to the planner, 

which will then return a plan annotated with a justifi-


cation for each plan step. The plan-implementation 

agent will then use the plan justification, together 

with further questioning of these agents, to discover 

why this failure occurred and which part of the theory 

is at fault. This problem point is then passed to the refinement 

system, which will then correct the problem 

and replace it with the correction in the original theory. 

This refinement process will not only allow us 

to achieve a goal that would otherwise be unreachable, 

but will also leave us with a domain theory that 

is richer and more accurate. 

Forming and Executing the Plan 

We need to find a plan for achieving the goal and to 

annotate this with a justification for each step. The 

reason this justification is required is that failure in 

the plan execution can be immediately linked to a 

problem in a particular part of the theory; namely, 

that part of the theory that was used to justify this plan 

step. We propose to use a state of the art planner such 

as FF so that our system is capable of producing long 

and complex plans if necessary. However, using such 

a planner will not provide us with information about 

the inference rules and justifications behind each plan 

step. Therefore we intend to build a plan deconstuctor 

that will take the plan produced by FF and, using 

the theory, pseudo-execute it, at each stage annotating 

the theory with the inference rule that was used 

and the preconditions of that rule, together with their 

justifications. 

Once the plan-implementation agent has received the 

annotated plan, it will attempt to execute it by interaction 

with other agents. For example, if the action 

required in a plan step is to buy a visa, it will need 

to locate an embassy agent who is capable of fulfilling 

this requirement. It is through these interactions 

that any fault in the theory will come to light. These 

agents will have their own internal theories about how 

to perform these actions, and these may not match the 

plan-implementation agent’s theory. 

For example, the plan-implementation agent may 

have a visa represented in his theory as a simple, 0- 


arity predicate, whereas the embassy agent may have 

it represented as a 1-arity predicate visa(country) 

which takes a country as an argument. Or perhaps the 

embassy agent has a more complicated rule for buying 

visas, which involves an extra precondition that 

the plan-implementation agent’s rule doesn’t have, 

for example, that an official invitation is required. In 

such cases, the actions will not be executed and the 

plan will fail. The plan-implementation agent can 

then question the embassy agent further, using the 

justification of the plan step, to find out exactly which 

part of the theory caused the failure and why. 

Refinement Techniques 

To create rules for specialising a theory or ontology, 

we looked first at rules for abstraction; that is, removing 

detail from a theory. We inverted these rules to 

form anti-abstractions which can thus be used to add 

detail to a theory: 

Predicate anti-abstraction - A single predicate is 

divided into some number of predicates 

Domain anti-abstraction - Constants and function 

symbols are divided up into different cases 

Propositional anti-abstraction - The arity of a 

predicate is increased 

Precondition anti-abstraction - Preconditions 

are added to rules. 

The refinement system will then select the appropriate 

type of refinement from those available to it, using 

the information gleaned earlier about the problem 

point. Once the refinement has been performed, it replaces 

the original in the theory, which can then be 

presented to the planner. 


Fiona McNeill, Alan Bundy, and Marco Schorlemmer 

Edinburgh University, UK, fionam@ 

dai.ed.ac.uk


ARTICLE 

MEXAR: Integrated AI Technologies to Support MARS EXPRESS 

Mission Planning 

Introduction 

Space exploration missions require coordination of a 

significant amount of activities. State of the art intelligent 

planning and scheduling (P&S) technology 

could potentially be of great help in supporting such a 

coordination. This work aim at showing an example 

of this technology in a support system for a specific 

mission scheduling problem related to the ESA program 

called MARS-EXPRESS [3]. 

MARS-EXPRESS is a space probe that will be 

launched during 2003 and after six months will be 

orbiting around Mars for the following two years and 

more. During the operational phase around Mars a 

team of people, the Mission Planners, will be responsible 

for the on board operations of MARS-EXPRESS. 

They receive input from different teams of scientists 

and cooperate with different specialists for various 

specific tasks (e.g., Flight Dynamics (FD) experts). 

Any single operation of a payload, named POR (Payload 

Operation Request), is decided well in advance 

through a negotiation phase among the different actors 

involved in the process (e.g., scientists, mission 

planners, FD experts). 

The result of our study is a system called MEXAR that 

addresses the memory dumping problem in MARS- 

EXPRESS. The specific problem that is addressed 

is defined as MEX-MDP and is described in detail 

in [7]. MEXAR is an interactive support system 

that may help mission planners in deciding policies 

for downlinking data to Earth during the temporal 

visibility windows. The tool uses constraint-based 

techniques for representing the basic problem to be 

solved, namely the segmentation of on-board memory 

in data packets during the visibility toward Earth. 

The paragraph below introduces the two solving 

algorithms which have been developed: a greedy 

Authors: G. Cortellessa, N. Policella, A. Cesta, and A. Oddi 

heuristic and a local search procedure [7, contains a 

complete explanation of these algorithms]. The following 

section describes an important aspect of this 

work, the interactive functionalities developed to support 

the user in his work [1, for more details]. 

The Packet Sequencing Algorithm 

Scheduling problems such as MEX-MDP can be seen 

as a special types of Constraint Satisfaction Problems 

(CSP) [6]. An instance of a CSP involves a 

set of variables ¡£¢¤ ¦¥¤§¨ �©�§��¤§¨ �� , a domain 

�� 

for each variable and a set of ��¡ 

constraints 

, � 

�� 

¥�� ©�� 

¢��¥¤§��©�§��§��¤� s.t. � 

which define feasible combinations of domain values. 

A solution is an assignment of domain values 

to all variables which is consistent with all imposed 

constraints. 

The CSP formalization of the MEX-MDP problem is 

based on a partition of � ¡ 

� � 

the temporal horizon 

in a set of � ¡�¢¤� ¥ ¡ 

� �� §�� contiguous � 

� windows � 

��¢¤��¡ � �¤��¥�§ ��¡ 

� 

§ � �� 

¥¨�� ¡ 

, such �� 

��§ 

� 

� ¡�� 

that . We 

�� 

�� consider 

a set of � � decision variables that 

¥ 

represent the 

amount of data to be dumped �� 

from packet store 

in the window �� . The MEX-MDP contains different 

kinds of constraints: (a) Given the characteristics of 

the packet stores the data must be downlink according 

to a FIFO philosophy; (b) the amount of data for each 

packet store does not exceed its capacity; (c) a finite 

amount of data can be dumped in each transmission 

window (finite transmission rate). 

All the proposed algorithms work over two levels of 

abstraction: (1) the planning level, where the whole 

set of decision variables are instantiated taking into 

account the different constraints; (2) the scheduling 

level, where a sequence of memory dump operations 

�


is generated over the communication links respecting 

the constraints imposed over all the windows � � . 

In order to find an optimal solution we choose to realize 

a heuristic optimization strategy based on local 

search which is able to improve an initial solution 

given as an input: Tabu search [4, 5]. The tabu metaheuristic 

is founded on the notion of a move. A move 

is a function which transforms one solution into another. 

For any solution ¡ 

, a subset of moves applied 

to ¡ 

is computed. The result is the neighborhood of 

¡ 

. The algorithm proceeds selecting, at each step, 

the best solution in the neighborhood, with respect to 

an objective function, till a fixed number of steps are 

made without finding better solutions. 

Figure 1: MEXAR layout 

In MEX-MDP the move consists in bringing forward 

some data (for example data contained in observations 

with the smallest volume of data) and delaying 

other ones; this should improve in many cases the objective 

function (mean turn over time). 

Mexar Interactive Functionalities 

The MEXAR functionalities that are designed for the 

users are summarized in Figure 2. As expected the 

problem solving activity is central in the system. This 

functionality is guaranteed by the automated services 

centered on the constraint-satisfaction methodology 

(CSP) described above. 

In the figure 2 we show the flow of control during 

the use of the functionalities. It is possible to iden- 


tify an activity that aims generically at defining a single 

problem. At present it simply consists of loading 

a problem description from a file, it can be also be 

replaced by a more complex incremental functionality 

that could be well coupled with the CSP modeling 

used. The definition of a problem is followed by 

its solution according to the different algorithms produced 

in our work. A different functionality allows 

to refine the current problem. This activity consists 

in deleting some of the Payload Operation Requests 

(PORs) from the associated timelines. This can be 

useful to experiment different loads on the resources 

in specific intervals of the solution. This functionality 

introduces a cycle among these activities that could 

bring the user to incrementally refine new MEX-MDP 

problems. As shown we group these functionalities 

in an interaction layout called Problem Analyzer (see 

Figure 2). 

Once a problem to solve is defined we can attack it 

with different solution methods. Figure 1 shows an 

example of a solved instance of MEX-MDP as it is 

presented to the user. 

Figure 2: MEXAR Interactive Environments 

The availability of a portfolio of problem solving 

procedures has suggested the idea of involving more 

deeply the user in the solution process. This has been 

pursued by creating an environment in which it is 

possible to save different solutions and, in addition, 

the user can guide search on how to improve the solutions 

applying different algorithms. We call this process 

solution space exploration. This aspect is strictly 

connected to the availability of evaluation metrics on 

the solutions as discussed in [2]. The idea behind the 

solution explorer is the one that the user can generate 

an initial solution, save it, try to improve it by local 

search, save the results, try to improve it by local


search with different tuning parameters and so on. In 

this way, it is possible to generate paths in the search 

space. The user can restore one of the previous solutions 

and try to improve it with a local search with 

different parameters, etc. In this way he generates a 

tree of solutions. This procedure can be repeated for 

different starting points generating, in this way, a set 

of trees. Using at the same time the evaluation capability 

on a single solution and its own experience 

he can generate different solution series, all of them 

saved, and, at the end, choose the best candidate for 

execution. 

Bibliography 

[1] Cesta, A.; Cortellessa, G.; Oddi, A.; and Policella, 

N. Interaction Services for Mission Planning 

in MARS EXPRESS. In Proceedings of the 

3rd International NASA Workshop on Planning 

and Scheduling for Space, 2002. 

[2] Cesta, A.; Oddi, A.; Cortellessa, G.; and Policella, 

N. Using AI Techniques to Solve MEX- 

MDP. Technical Report MEXAR-TR-02-08, 

ISTC-CNR - Planning and Scheduling Team, 

Rome, Italy, 2002. 

[3] ESA. European Space Agency, MARS 

EXPRESS Web Site. http://sci.esa.int/ 

marsexpress/, 2002. 

[4] Glover, F. Tabu Search – Part I. ORSA Journal 

of Computing 1:190–206, 1989. 

[5] Glover, F. Tabu Search – Part II. ORSA Journal 

of Computing 2:4–32, 1990. 

[6] Montanari, U. Networks of Constraints: Fundamental 

Properties and Applications to Picture 

Processing. Information Sciences 7:95–132, 

1974 

[7] Oddi, A.; Cesta, A.; Policella, N.; and Cortellessa, 

G. Scheduling Downlink Operations in 

MARS EXPRESS. In Proceedings of the 3rd 

International NASA Workshop on Planning and 

Scheduling for Space, 2002. 


Gabriella Cortellessa ¥ , Nicola Policella © , 

Amedeo Cesta, and Angelo Oddi ISTC- 

CNR, Italian National Research Council, Viale Marx 

15, I-00137 Rome, Italy, {corte,policella, 

cesta,oddi}@ip.rm.cnr.it 

1 is also Ph.D. student in Cognitive Psychology at 

University of Rome “La Sapienza” 

2 is also Ph.D. student in Computer Science at University 

of Rome “La Sapienza” 

RDPPlan: an Extension of DPPlan for Planning with Interval 

Resources 

RDPPLan is a model for planning with quantitative 

resources. It is based on DPPPlan [1], a planner 

which uses a non directional search algorithm on the 

planning graph. 

Most models of planning with resources, like [3], [4], 

[5], [6], [7] and [8], assume that an exact value can 

model the continuous quantities describing, in the 

real world, a given resource. In other words these 

ARTICLE 

Authors: M. Baioletti, A. Milani, and V. Poggioni 

models cannot deal with more realistic situations in 

which quantities are not known exactly. The RDP- 

Plan model allows one to manage domains where preconditions 

and effects over quantitative resources can 

be specified by intervals of values; in addition mixed 

logical/quantitative and pure numerical goals can be 

specified. 

Instead of initializing a resource with a unique real


¢¡ 

£ 

£ � � 

¤¥¡§¦ ¡¨¦ 

value, we allow for the specification of a real interval 

as the range of the initial value of the resource 

. The planner operates in an underspecified domain 

in which the value of some resource is not exactly 

known, but it is bound to be in an interval. Let us suppose 

that we do not know exactly how much gasoline 

is in the tank of our car: we just know for sure that the 

real amount is between 5 and 10 liters. Similarly it is 

possible to have underspecified effects of some operators: 

the value which is added (subtracted), multiplied 

(divided) or assigned to the current value of a 

resource is not exactly known, but only a lower and 

an upper bound is specified. Let us imagine that we 

do not know which is the exact consumption of the 

car in the previous example: all we know is that the 

car travels from 10 to 15 kilometers per liter. 

We define two intervals associated to each resource 

and each time level : and that respectivly 

represent the Realized Interval and the Desired Interval 

of resource £ at time-step � ¤©¡§¦ 

¡ 

. is initialized with 

and it is updated, at each step, by the effects on 

the resource £ of the actions already inserted in the 

plan; � 

¡¨¦ , instead, contains all the admissible values 

for the resource £ 

that allow the execution of all the 

actions selected at time level � . 

In this model the necessary and sufficient condition 

for a plan to be a solution of a given planning problem 

is that for each resource £ 

and time level � � 

¡¨¦ ¤�¡§¦�� 

¥ � © 

� 

§�� 

� 

§�� 

�� 

§�� 

�� 

¥ � � 

, 

. In other terms, a solution plan must 

solve every possible problem that is allowed by the 

constraints specified in the initial state and in the effects 

description. 

We use the same concept of simultaneous executability 

as expressed in [2] and in [6]. We say that 

the actions are simultaneously executable 

if, for every permutation of , each 

action is executable in the order defined by the permutation 

and the effect over resources is always the 

same. As a straightforward consequence, an assignment 

on resource £ 

is not simultaneously executable 

with any action changing £ � 

correct method that can allow us to calculate the Desired 

Interval for the simultaneity execution of actions 

that have preconditions and/or effects on the 

same resource 

and an additive operator 

(increase, decrease) is not simultaneously executable 

with any multiplicative operator (scale–up, 

scale–down). Moreover we have defined a provably 

£ �� 

¡§¦ ¤�¡§¦ 

. 

In order to achieve the goals over resources, we 

have defined strategies to solve ”pure numerical problems“, 

i.e. with goals only on resources. Such strategies 

are combined with the ones for solving logical 

goals, by evaluating the difficulties of resources and 

logical goals and selecting the most difficult goal to 

solve. 

When we work with resources as intervals, we have 

to handle with real intervals whose width in general 

grows, except when an assignment is performed. 

Moreover note that if the width of the Realized Interval 

is large, it is more difficult that the solution 

condition holds. 

The action choice criterium takes into account the 

distance between the corresponding Desired and Realized 

intervals and their width. In particular we 

have defined two preference functions, one for interval 

widths, and one for distance between intervals, 

and the algorithm chooses the action that maximizes 

a linear combination of these functions. 

Investigations and experiments are planned in order 

to develop more accurate heuristics and strategies 

which take resources into account. Moreover, in order 

to provide a meaningful evaluation, it will also 

be required the development of a set of significant 

benchmarks for planning domains with interval resources. 

Finally it is worth investigating further extensions 

to the resource model more accurate with respect 

to the uncertainty in the real world e.g. intervals 

with given probability distribution over resource values 

and fuzzy quantities. 

Bibliography 

[1] Baioletti M., Marcugini S. and Milani A. DP- 

Plan: an Algorithm for Fast Solution Extraction 

from a Planning Graph, In Proc. of AIPS-2000 

[2] Fox M. and Long D., PDDL 2.1: An Extension 

to PDDL for Expressing Temporal Planning Domains, 

2002 

http://www.planet-noe.org


[3] Koehler J. Planning under resource constraints. 

In Proc. of ECAI-1998 

[4] Chien S. et al., ASPEN Automated Planning 

and Scheduling for Space Mission Operation, In 

Proc. of SpaceOps 2000 

[5] Wolfman S. and Weld D., The LPSAT Engine and 

its Application to Resource Planning, In Proc. of 

IJCAI 1995 

[6] Rintanen J. and Jungholt H. Numeric State Variables 

in Constraint-Based Planning, In Proc. of 

ECP-1999 

[7] Currie K. and Tate A., O-Plan: The open planning 

architecture, In Artificial Intelligence 52, 

49-86, 1991 

[8] Laborie P. and Ghallab M. Planning with 

sharable resource constraints, In Proc. of IJCAI 

1995 


Marco Baioletti Dip. Met. Quantitativi, Università 

di Siena, Italy 

Alfredo Milani Dip. Mat. e Informatica, Università 

di Perugia, Italy 

Valentina Poggioni Dip. Informatica e Aut., 

Università di Roma3, Italy 

ARTICLE 

Extending Operator Induction to Provide Full Method Sets for 

Hierarchical Planning Domains 

Abstract Modelling a world for efficient HTN planning 

by hand is a lengthy and time consuming process. 

Our knowledge acquisition tool, GIPO (Graphical 

User Interface for Planning with Objects), offers 

GUI abstraction allowing the user to concentrate on 

the model rather than a language syntax. Using GIPO 

we aim to provide automatic operator induction for 

planning domains with an hierarchical sort structure 

so that complex hierarchical operators can be constructed. 

At present we can induce operators for flat domains. 

Using GIPO we interactively construct a sequence of 

actions to arrive at some goal state. Then we input 

a partial domain (with no operators) and using opmaker, 

the induction tool in GIPO, we obtain a set 

of operators to complete the constructed task. These 

operators may not be accurate at this stage. 

We can also induce operator sequences to construct 

low level methods (hierarchical operators for HTN 

domains). The name method implies that these are 

different recipies for achieving similar or related 

tasks and as such they often repeat actions or ac- 

Author: N.E. Richardson 

tion sequences. Issues arising from using repetition 

in the sequences are that the same operators are induced 

more than once and, as it is desirable to have 

only one operator per action, we can use repetition to 

generalise the operator. The debate is then to find a 

way of revising the operator each time a new version 

is induced or employ a learning and revision process. 

We propose two systems to compare induced operator 

specifications and generalise those descriptions. 

Comparing operators will give us a category of differences. 

Some differences are allowable in the present 

system. For example an operator can be induced 

without a conditional clause and if the same operator 

is used later in the sequence with a condition then 

the conditional clause is added to the original operator. 

We would want to be able to merge operators 

that have different names but are otherwise identical. 

Some operators will be more general than others but 

we recognise that over-generalisation can mean that 

the operator is not sufficiently expressive. 

Operators describe objects’ transitions from one set 

of substates prior to the action represented, to another


set of substates after the action. In generalising an operator 

the aim is to make it applicable to a wider set 

of objects but this can be overdone. A difficulty in 

generalising operators is to limit extent that the operator’s 

expressiveness is compromised. 

Generalising operators and revising the operators in 

the system as new ones are generated will allow us 

to constuct the higher levels of the method hierarchy 

which are built from other methods as well as primitive 

operators. When we have these systems in place 

The SimPlanner 

Introduction 

Off-line planning generates a complete plan before 

any action starts its execution [3]. This forces to 

make some assumptions that are not possible in real 

environments like, for example, that actions are uninterruptable, 

that their effects are deterministic, that 

the planner has complete knowledge of the world or 

that the world only changes through the execution 

of actions. On the other hand, on-line planning allows 

to start execution while the planner continues 

working in order to improve the overall planning and 

execution time. Nowadays there are only some few 

approaches for planning in dynamic environments 

and/or with incomplete information [2]: 

Conditional planning: this approach tries to consider 

the possible contingencies that can occur in 

the world. 

Parallel planning and execution: this approach 

separates the planning process from the execution. 

Interleaving planning and execution: this approach 

allows quick and effective responses to 

changes in the environment. 

SimPlanner is an integrated tool for planning and execution, 

and it is based on this latter approach. This 

tool is thought to be used in real environments such 


it will be possible to have full method induction in 

GIPO for hierarchical domains. 


N.E. Richardson School of Computing and Engineering 

The University of Huddersfield, Huddersfield 

HD1 3DH, UK, n.e.richardson@hud. 

ac.uk 

ARTICLE 

Authors: O. Sapena and E. Onaindía 

as the intelligent control of robots. However, it has 

initially been implemented as a simulator in order to 

check its behavior without having to integrate it in 

different several domains. SimPlanner is made up of 

three components: an on-line planner, a monitoring 

module and a replanner. 

The on-line planner 

The on-line planner is responsible for generating, in 

an incremental way, a plan to achieve the goals. As 

soon as the planner calculates the first action, the plan 

can start its execution. From this moment on, the 

planning and execution processes keep on working in 

parallel while no unexpected event is detected; otherwise, 

the execution must wait for the planner to make 

the necessary modifications in the plan. 

The planner is based on a depth-first search, with no 

provision for backup. The planning decisions (inferred 

actions) are consequently irrevocable. The 

planning algorithm uses heuristic functions to compute 

an approximate plan (AP) for each goal independently. 

Then, a conflict-checking mechanism detects 

conflicts among actions in the approximate plans and 

selects the action from the AP that minimizes the 

number of conflicts. The selected action is inserted 

at the end of the plan, and the current state is updated 

through its execution. This algorithm is iteratively


executed until all top-level goals are achieved [4]. 

The monitoring module 

Monitoring is the process of observing the world and 

trying to find discrepancies between the physical reality 

and the beliefs of the planner [1]. Basically, there 

exists two types of plan execution monitoring [2]: action 

monitoring checks the validity of the action preconditions 

before it starts its execution and also that 

its effects have taken place as expected. The environment 

monitoring tries to capture information from the 

external world that conditions the remaining planning 

process. Monitoring is, therefore, domain-dependent. 

Since SimPlanner is being used at the moment as a 

simulator, this information is input to the system by 

the user. The user is who decides what information 

the robot receives and which unexpected events that 

occur in the world are communicated. 

The replanner 

When an unexpected event is detected, the calculated 

plan is checked in order to assure it is still valid [1]. If 

this is the case, the execution simply continues. Otherwise 

the replanning module is invoked. The replanner 

tries to reuse as much of the calculated plan as 

possible without losing the quality of the final plan. 

It uses a heuristic function to find out which is the 

best reachable state through the actions in the original 

plan. If there are many reusable actions, the planning 

process will save a lot of computation time. In 

the worst case, a new plan will be computed from 

scratch. The replanner overhead is very small so it 

is worth trying to reuse the plan rather than planning 

from scratch [5]. 

Bibliography 

[1] G. De Giacomo and R. Reiter, ‘Execution monitoring 

of high-level robot programs’, Principles 

of Knowledge Representation and Reasoning, 

453–465, (1998). 

[2] K.Z. Haigh and M. Veloso, ‘Interleaving planning 

and robot execution for asynchronous user 

requests’, Papers from the AAAI Spring Symposium, 

35–44, (1996). 

[3] M.E. Pollack and J.F. Horty, ‘There’s more to 

life than making plans: Plan management in dynamic, 

multi-agent environments’, AI Magazine, 

20(4), 71–84, (1999). 

[4] O Sapena and E. Onaindia, ‘Domainindependent 

on-line planning for STRIPS 

domains’, Proceedings of Iberamia-02, LNAI, 

2527, 825–834, (2002). 

[5] O Sapena and E. Onaindia, ‘Execution, monitoring 

and replanning in dynamic environments’, 

Workshop on On-Line Planning and Scheduling, 

AIPS-02, (2002). 


Oscar Sapena, Eva Onaindía Departamento 

de Sistemas Informáticos y Computación, Universidad 

Politécnica de Valencia, Spain, {osapena, 

onaindia}@dsic.upv.es


Answer Set Planning with DLV 

Introduction 

The knowledge based planning system DLV implements 

answer set planning on top of the DLV system 

[1]. It is developed at TU Wien and supports the 

declarative language ¥ 

[3] and its extension ¥¢¡ [4]. 

The language ¥£¡ 

is syntactically similar to the action 

language £ 

¡ ¥ 

[6], but semantically closer to answer set 

programming (by including default negation, for example).tures: 

offers the following distinguishing fea- 

- Handling of incomplete knowledge: for a fluent ¤ , 

neither ¤ nor its opposite ¥¦¤ need to be known in 

any state. 

- Nondeterministic effects: actions may have multiple 

possible outcomes. 

- Optimistic and secure (conformant) planning: 

construction of a “credulous” plan or a “sceptical” 

plan, which works in all cases. 

- Parallel actions: More than one action may be executed 

simultaneously. 

- Optimal cost planning: In ¥ ¡ , one can assign an 

arbitrary cost function to each action, where the 

total costs of the plan are minimized. 

An operational prototype of DLV as well as sample 

encodings of planning domains in the system are 

available at http://www.dlvsystem.com/K/. 

© 

input 

Background 

Knowledge 

Control Flow 

Data Flow 


© Parser 

Datalog Parser 

Underlying Concepts 

ARTICLE 

Author: A. Polleres 

¥ ¡ ¡ ¥ 

at(§ 

at(� travel(��§¨§ 

Action language In transitions are 

described in a declarative way by means of 

causation rules, e.g. caused ) after 

), ). Furthermore, the language 

offers constructs to express executability and 

non-executability of actions, ramifications, nondeterminism 

and allows to assign costs to actions. 

Planning problems in this language are transformed 

to a logic program which is then evaluated under the 

answer set semantics. 

Answer Set Programming Answer Set programs 

are logic programs in a syntax similar to Prolog 

which allow for negation as failure in rule bodies 

and disjunction in rule heads. In our approach, 

the minimal models of these programs under the so 

called Answer Set Semantics [5] correspond oneto-one 

to the plans of the resp. planning problem 

specified in ¥ ¡ . This view of models representing 

plans can be partly compared to planning using SAT- 

Solvers. 

System architecture 

The architecture of DLV is outlined in Figure 1. The 

input of the system consists of domain descriptions 

(DLV files) and optional static background knowledge 

specified by a logic program. 

Plan Generator 

DLV © Core 

DLV Core 

Figure 1: DLV © System Architecture 

Plan Checker 

Controller 

Plan Printer


The Controller first invokes the Plan Generator, 

which translates the planning problem at hand into 

a suitable program in the core language of DLV (disjunctive 

logic programs under the answer set semantics). 

Then, the DLV kernel is invoked to solve the 

corresponding problem. The resulting answer sets (if 

any exist) are fed back to the Controller, which extracts 

the solutions to the original problem from these 

answer sets and transforms them back to the original 

planning domain. 

If the user specified that secure/conformant planning 

should be performed, the Controller then invokes the 

Plan Checker which verifies by another evaluation of 

a logic program whether this plan is in fact also secure/conformant. 

Finally, the solutions found by the Generator (and optionally 

verified by the Checker) are translated back 

into user output and printed. 

Details about the transformations mentioned above 

can be found in [2]. Performance and experimental 

results are reported in [2, 4]. 

Bibliography 

[1] T. Eiter, W. Faber, N. Leone, and G. Pfeifer. 

Declarative Problem-Solving Using the DLV 

System. Logic-Based Artificial Intelligence, pp. 

79–103. Kluwer, 2000. 

[2] T. Eiter, W. Faber, N. Leone, G. Pfeifer, and 

A. Polleres. A Logic Programming Approach to 

Knowledge-State Planning, II: the DLV System. 

Technical Report, December 2001. To appear in 

Artificial Intelligence. 


A. Polleres. A Logic Programming Approach to 

Knowledge-State Planning: Semantics and Complexity. 

Technical Report, December 2001. To 

appear in ACM Transactions on Computational 

Logic. 


A. Polleres. Answer Set Planning under Action 

Costs. Technical Report, October 2002. 

[5] M. Gelfond and V. Lifschitz. Classical Negation 

in Logic Programs and Disjunctive Databases. 

New Generation Computing, 9:365–385, 1991. 

[6] E. Giunchiglia and V. Lifschitz. An Action Language 

Based on Causal Explanation: Preliminary 

Report. In AAAI ’98, pp. 623–630, 1998. 


Axel Polleres Institut für Informationssysteme, 

TU Wien, A-1040 Wien, Austria, axel@kr. 

tuwien.ac.at, pfeifer@dbai.tuwien. 

ac.at 

This work was supported by FWF (Austrian Science 

Funds) under the projects P14781 and Z29-INF. 

Furthermore, the author thanks the ÖGAI (Austrian 

Association for Artificial Intelligence) and the OCG 

(Austrian Computer Society) for sponsoring travel 

costs to the PLANET Summer School 2002


Planform: An Open Environment for Building Planners 

Project Overview 

The PLANFORM project aimed to develop a high level 

research platform for the systematic construction of 

planner domain models and automatically configured 

planning algorithms. 

Our objectives were, briefly: 

To assemble tools within an open environment for 

the acquisition and modelling of planning domain 

models; 

To create languages for modelling of planning domains 

and to specify characteristics of planners 

leading to the configuration of purpose-built planning 

engines; 

To create a tool which synthesises a domain model 

and a planning engine into a planning application; 

To evaluate the approach using realistic problem 

domains. 

During the project, the emphasis changed slightly so 

that some of objectives developed in a different direc- 

Application 

Domain 

Knowledge 

Acquisition 

Tools 


Internal 

Representation 

of 

Domain 

Static and Dynamic Analysis 

ARTICLE 

Authors: T.L. McCluskey, M. Fox, and R. Aylett 

tion. Rather than abstractly defining planning algorithms, 

we decided to create a library of algorithms 

and use domain analysis technology to design and 

configure a planning application. We perceived the 

knowledge acquisition bottleneck to be a significant 

problem for AI Planning and concentrated more resource 

than planned on this area. 

We have made substantial progress towards our objectives. 

We have developed an environment enabling 

the acquisitition and modelling of planning applications 

and the configuration of planning engines suited 

to those applications. Although time limitations prevented 

us from completing all aspects of the final integration 

and evaluation we achieved our main objectives 

and laid a strong foundation for development of 

the project. 

The project consisted of a model engineering phase 

and a planner engineering phase. Huddersfield 

and Salford collaborated closely on modelling and 

knowledge acquisition issues, whilst Huddersfield 

and Durham collaborated on the development of parts 

of the domain modelling tools. Durham worked on 

the planner engineering phase of the project. 

Configuration 

Tool 

General 

Planners 

Generic 

Heuristics 

Model Engineering Phase Planner Engieenring Phase 

Figure 1: Architectural Breakdown of Planform 

Efficient 

Planning 

Application


Research at the University of Huddersfield 

The Huddersfield contribution emphasised tools creation 

and integration aspects of the project. Tools for 

acquisition, validation and domain modelling were 

developed, in collaboration with colleagues at both 

Salford and Durham. As part of this effort the Huddersfield 

team researched and developed a GUI-based 

environment called GIPO 1 . In addition, as the initiators 

of PLANFORM, Huddersfield performed a central 

administrative function, producing the main external 

website and also an internal website with additional 

resources such as Project Meeting minutes. 

NB: in the text below ‘the PLANFORM website’ refers 

to http://scom.hud.ac.uk/planform. 

The first phase of the Huddersfield contribution was 

concerned with applications encoding and language 

development. During this phase three application domains 

were used to explore the adequacy of the modelling 

languages ��¢¡ and ��¢¡¤£ [20, 19]. ��¥¡¦£ 

is a hierarchical version of ��¢¡ enabling the modelling 

of planning domains as hierarchical task decompositions. 

Salford and Huddersfield collaborated 

on this effort with Salford working on the modelling 

of some existing domains using the two languages. 

Huddersfield tackled the problem of encoding the aircraft 

landing scheduling problem supplied by Mark 

Watson of the National Air Traffic Services. The constraints 

(e.g. separation times for each type of aircraft) 

were encoded in ��¢¡ [18] and ��¢¡§£ . ��¢¡¦£ 

was found adequate for all three domains, although 

the encoding did demonstrate some required changes, 

and overall the pressing need for tool support. 

The standard languages for communicating planning 

domain descriptions are the PDDL variants [1]. In order 

to be able to experiment with our domain models 

using exisiting planning technology we therefore 

created tools to map between ��¢¡ and PDDL. 

These tools continued to be developed throughout the 

project allowing planners, and other tools that receive 

input in PDDL form, to be integrated into our environ- 

1 http://helios.hud.ac.uk/planform/gipo 

ment. There are some interesting technical aspects of 

the mapping discussed in [27]. 

Language manuals for ��¢¡ and ��¥¡¤£ were maintained 

during the project [12] and an online help facility 

was constructed (see the PLANFORM website). 

Drawing on previous development work (e.g. [19]), 

we assembled tools that automate the syntactic and 

semantic analysis of ��¥¡ domain models. Analyses 

include ensuring that invariant properties of the 

model are maintained and that syntactic rules are observed. 

The second phase of the project was concerned with 

development of the GUI and tools environment. The 

focus was on building and integrating knowledge acquisition 

and modelling tools for AI planning into an 

open environment. The GUI and some of the tools 

described above were built in Java. Others were implemented 

in Prolog and it was necessary to integrate 

these via Sicstus Prolog’s JASPER interface. 

The JavaCup parser generator method was used to 

represent the syntax rules of ��¢¡ and to generate 

a parser for the language. This formed the input tool 

in GIPO (Graphical Interface for Planning with Objects). 

The knowledge acquisition part of the tool 

was structured using the method outlined in earlier 

work [20]. The methods direct the user to define objects, 

object sorts, relations and properties, classes 

and constraints on object situations, problems in the 

form of task specifications, and finally operators built 

from these components. The design of the interface 

was based on the need to minimise the use of syntax, 

and use object rather than predicate centred ideas. 

Once users have entered all parts of a domain model 

they can utilise modelling tools to remove bugs and 

experiment with the encoding. We created and 

adapted the following tools. 

plan stepper: This allows the user to pick action 

schemas and apply them to a state, until a desired 

goal is reached. It is useful for identifying errors 

in operators and operators sets. 

internal planning engines: this allows our own 

in-house planning engines to be connected up to


GIPO. Sample tasks can be executed and the resulting 

solutions displayed. 

interface for external planning engines: This allows 

external planning engines to be ‘bolted’ into 

the environment. The planner needs to be able to 

input domain models in PDDL (from GIPO), and 

output solution in a prescribed format. Again, the 

resulting solutions are displayed through GIPO. 

a random task generator: This inputs the current 

domain model and randomly generates tasks to be 

used with a planner. 

an animator: After a domain model has been entered, 

and the planning engine has solved a task 

within that model. the animator can be used to 

track the transitions of each of the objects which 

started in the initial state. 

In the third phase, integration and evaluation, the 

tools outlined above were integrated into GIPO [28]. 

The software was released and demonstrated at 

ECP’01, and again at AIPS’02. It is available on 

Unix, Linux and Windows platforms from the PLAN- 

FORM website. As an initial indication of GIPO’s 

impact, the Huddersfield website recorded 147 external 

downloads of the system in the period November 

2001 - March 2002. 

The environment has been tested using a range of 

common domains (details are in the resource section 

of the website). Further, GIPO was used as a teaching 

tool in a second year introductory course in Artficial 

Intelligence. GIPO alleviates many user interface 

problems by adopting an object modelling approach 

which seems natural to non-expert users. To ameliorate 

the use of GIPO by non specialists the following 

issues were explored: 

The use of an inductive approach to capturing operators 

2 . The opmaker tool was created which 

outputs a set of operators given a partial domain 

model and an example solution sequence [22, 21]. 

The use of generic types to suggest planning design 

patterns. These ideas were developed as 

2 This work was primarily work undertaken in conjunction with a PhD student 


part of the Durham PLANFORM project, and Huddersfield 

is working in collaboration with colleagues 

at Durham to integrate design patterns 

into GIPO [26]. 

The final phase of our work has been to extend the internal 

language and the surrounding tools from ��¥¡ 

(version 1) to ��¥¡¦£ (version 2). ��¥¡¦£ extends 

��¢¡ in two major ways: HTN operators can be 

used, and sort constraints can be put on each level 

of the sort hierarchy meaning that objects of a primitive 

sort inherit all the constraints up the hierarchy. 

This modelling approach is being tested using a large 

‘Translog’ domain imported from a transport logistic 

domain constructed by members of the University of 

Maryland. 

The priorities for future development of the Huddersfield 

contribution are: 

The development of a suite of planning design 

patterns and their integration with the GIPO tool; 

�©¨ ¨ 

The �¡ £¢¥¤§¦ 

£ 

evolution of the tool into a general 

mixed-initiative plan authoring tool; 

Integration of the operator induction techniques 

with a plan authoring interface so that operator 

specifications can be induced and refined interactively. 

The development of the ��¥¡ representation language 

to be on the expressive level of PDDL2.1 

(temporal representations), thereby enabling GIPO 

to support the modelling of temporal planning domains. 

Research at the University of Salford 

Part of the set of objectives for the PLANFORM 

project was to make AI planning technology accessible 

to non-experts. In pursuit of this objective, work 

at Salford was based on the idea of the Knowledge 

Level [25] and Models of Expertise [6] as articulated 

over many years in the KBS community, in which the 

problem level is modelled separately from the design 

� �


level. Research in KBS technology has shown that 

support for domain experts is feasible if it is based on 

the generic task concept [7], and much earlier work 

has been carried out round the generic task of diagnosis 

[24]. Oddly, little of this has been applied to 

AI Planning. The basic idea is that a generic task 

incorporates a skeletal model at the knowledge-level 

which can then be used to direct a computer-based 

knowledge acquisition process with a domain expert. 

Thus, to support domain experts, it was seen as necessary 

to build a knowledge-level tool as part of the 

PLANFORM environment, incorporating generic task 

components for planning, and supporting knowledgelevel 

construction of the planning domain rather than 

forcing the use of the domain design language, ��¥¡ , 

used internally. However such a tool would necessarily 

link to the ��¢¡ GIPO tools being developed 

at Huddersfield (see Figure 1) and thus output 

��¥¡ , so it was therefore vital to understand how 

planning problems would be represented in ��¥¡ . A 

strong link with Huddersfield was built through modelling 

activity in which the objective was to understand 

this mapping and derive some constraints on 

the knowledge-level interface, which was oriented towards 

users with no AI planning knowledge but with 

expertise in a particular planning domain. A number 

of domains were modelled including the multi-robot 

Drumstore from earlier work at Salford. 

Planning ontologies were identified as a key foundation 

for such a knowledge-level tool, and the means 

by which the skeleton model mentioned above could 

be embodied. A survey of work in the field was carried 

out and the possibility of incorporating an existing 

ontology such as CYC( http://www.cyc.com/cyc- 

2-1/cover.html.) was investigated but limited time did 

not allow its use. 

The third component researched as the basis for a 

knowledge-level tool was the requirements of the domain 

expert, and here an accessible domain was formulated 

(EVENTUS – arranging a weekend outing for 

a visiting researcher) and a knowledge acquisition exercise 

was carried out with four people. The exercise 

was repeated with the robot Drumstore problem. 

Data was analysed for concept coverage and for in- 

terface design issues, looking at the process a human 

expert goes through in conceptualising the domain. 

The details of this process are described in [2]. 

The KA tool was constructed in Java, and in its first 

version supported passive model construction by the 

expert with support from a small hand-coded ontology 

for the domains already investigated. In its second 

version, an active question-driven process was 

added based on the key planning concepts of Task, 

Agent and Object [5]. 

The methodology embodies two successive 

extraction-refinement processes: protocol to problem 

specification; and problem-specification to conceptualisation. 

A part of the KOD (Knowledge Oriented 

Design) method [30] was applied to obtain an accurate 

process for knowledge acquisition and to build 

the conceptual model through a set of examples and 

scenarios. 

The output of the KA-Tool is ��¢¡ , which can then 

be loaded into the GIPO tool created at Huddersfield. 

By the formal end of the project, it was possible to 

generate the world model in ��¢¡ , and since then it 

has become possible to generate planning operators, 

seen by most people as a key problem in formulating 

a planning domain description. In the 26 months of 

the project, it was not possible to carry out any extensive 

evaluation programme, but it is proposed to carry 

on with the work for a limited period informally with 

this as the key task. 

The modelling exercise enabled the development of 

strong links with Huddersfield, where frequent (almost 

weekly) visits were made at some points. A 

two-week visit to Durham was also organised to 

strengthen understanding of the requirements of the 

generative planning back-end. 

Further development of the KA-Tool is still being carried 

out in house, with the generation of Planning Operators 

now possible. If sufficient resources can be 

found, the next steps would include: 

Incorporation of a large ontology, such as CYC, 

into the KA tool; 

Integration of Durham’s generic types into this ontology;


Full integration of the KA tool with GIPO. 

Two publications were generated jointly with the 

Huddersfield team (Simpson et al 2001a, 2001b) 

and two more by the Salford team on their own [4] 

and [2]. One further paper is under review [3] and a 

journal article is in preparation. 

Research at the University of Durham 

The Durham contribution to PLANFORM focussed on 

the development of planner configuration technology. 

The objective was to develop techniques by which, 

given a domain model elicited from a user, planner 

components suited to the domain could be automatically 

identified and configured into a purpose-built 

planning system. Our approach has been to maintain 

a library of planner components, including heuristics, 

specialised solution strategies and problem-specific 

control rules, and to access these by means of patternmatching 

techniques once patterns have been identified 

in the domain model. 

The techniques used to identify domain patterns are 

based on static domain analysis algorithms developed 

at Durham prior to the start of the PLANFORM 

project [8]. The objective in the project was to extend 

these algorithms to enable the recognition of generic 

types and associated patterns of behaviour in planning 

domains, and to associate these generic patterns 

with special purpose solution strategies [14, 13, 16]. 

Briefly, a generic type is a collection of types sharing 

some fundamental behaviour. For example, the 

generic type of mobility contains all types of objects 

that are capable of movement while the generic type 

of construction contains all types that are capable of 

being combined into compounds (and subsequently 

recovered by destruction of the compound). 

When a generic type is present its associated patterns 

of behaviour are present and these can be used both 

to assist a domain designer in refining the model and 

to suggest appropriate solution approaches. 

The analysis techniques developed at Durham can 

identify certain key generic types and associated patterns. 

For example, the pattern associated with mo- 


bility comprises the mobile types, their maps and the 

predicate that defines locatedness of a mobile object 

on its map. A specific problem associated with mobility 

is route-planning, and a special purpose solution 

strategy suited to this problem can be to exploit 

travelling saleman heuristics. We were able to automate 

the configuration of planners with specialised 

route-planning capabilities enabling route-planning 

sub-problems to be handled using specialised approaches 

instead of by search [9, 13, 10]. 

The following software was developed at Durham as 

part of the PLANFORM project: 

Versions 4 and 5 of the STAN planning system. 

STAN [9, 10] performs generic type analysis and 

configures a special purpose planner suited to the 

associated generic patterns; 

Extensions to TIM [8, 15] to recognise a range of 

generic types in a domain model; 

The TIM API providing access to the generic type 

analyses performed by the TIM system, enabling 

their exploitation by other planning systems. The 

API is being exploited by other researchers in the 

international community; 

The OODL domain modelling language, supporting 

the construction of domains around generic 

types, and associated domain modelling tool 

DRAUGHTSMAN. These were developed by an 

MSc student and contributed to our collaboration 

with Huddersfield on the GIPO tool. 

In the scope of the project a number of other generic 

types have been identified (for example, construction, 

resource-allocation, portability and others) and associated 

with specialised solution strategies. The configuration 

problem becomes complex when several 

generic patterns co-occur and their solution strategies 

must be integrated, and we have not completed 

the work required to support arbitrarily complex patterns 

of integration. We have, however, categorised 

the forms of integration that need to be handled and 

made progress with configurations based on several 

of these categories [17, 11, 9].


The library contains stored parametrized solution 

strategies (such as travelling salesman solvers, multiprocessor 

scheduling heuristics etc) appropriate for 

sub-problems that commonly arise in planning domains. 

We do not try to guarantee complete coverage 

of all such sub-problems – the configuration system 

defaults to search if no generic types or suitable 

library components can be found. At present the library 

contains only one solution strategy per identified 

generic type, so extraction of a suitable stategy 

is simple. In general the extraction problem is 

more difficult because there may be different forms 

in which generic patterns arise and these might need 

to be matched in some intelligent way against the library. 

We have not explored this issue in the scope 

of PLANFORM. A recent extension we have made to 

the library is the addition of generalised control rules 

which can be selected and instantiated to fit a specific 

problem domain [23]. 

Most recently our work in these areas has considered 

the use of generic patterns as a basis for the development 

of planning design patterns. Using these, a 

domain construction tool can prompt the user for the 

components of generic patterns in a way that makes 

it simple for the user to enter that information. Initial 

work on a tool capable of supporting this idea was 

done by an MSc student [29] who was temporarily 

employed on the PLANFORM project at Durham. The 

work was continued in collaboration with the Huddersfield 

site which has focussed on the development 

of tool support [26]. 

The planner configuration approach has been tested 

by entering a hybrid planning system, STAN version 

4, into the international planning competition 

in 2000. STAN 4 can automatically detect mobility 

and resource-dependence patterns in planning domains 

and can extract route-planning and resourceallocation 

strategies from its library. Selected strategies 

are integrated, by means of a simple constraintbased 

interface, to a forward-search-based planner 

[10]. STAN 4 was one of the prize-winning systems, 

selected for the promise it showed in utilizing 

novel approaches to solving complex planning problems. 

Its plan quality was generally superior to that 

produced by the other competing systems. 

Work remains to be done on increasing the sophistication 

of the integration techniques that support coordination 

of different specialised solution strategies 

within the overall framework. An important aspect 

of making the configuration tools available to the 

general planning community is to provide a clean 

means of access to the library, pattern recognition 

techniques and planner interface. A priority for future 

development is to supply an API to the suite of 

tools we have, which other planning systems could 

exploit. Although we made progress with the design 

and implementation of such an API we did not complete 

its implementation and it remains a topic for future 

work. 

Overall Conclusions 

The PLANFORM project set out to construct an 

open environment for planner development, bringing 

knowledge acquisition tools, domain construction 

tools, modelling languages and planner configuration 

components into an integrated organisation making 

planning accessible to the non-specialist. 

The domain construction tool developed at Huddersfield 

produces PDDL domain descriptions providing 

a simple connection to the planner configuration architecture 

developed at Durham. The knowledge acquisition 

tools developed at Salford assist a user in 

confronting the task of domain construction through 

the GIPO tool. At this stage it is possible for a naive 

user to follow the entire process of modelling, and 

planning with, a specific problem domain without requiring 

detailed knowledge of any internal representation 

language ( ��¥¡ or PDDL). The level of abstraction 

at which such a user can work within the 

environment will be further raised when the implementation 

of design patterns, as a guiding principle 

in the modelling process, is completed. 

More time and work is necessary to evaluate the environment 

in terms of how successful it is at making 

planning accessible to the non-expert. There are several 

key aspects of the environment that require evaluation:


The extent to which the object-oriented approach 

to modelling ameliorates the modelling task for 

the planning non specialist; 

The extent to which design patterns can further 

ameliorate this effort; 

The extent to which the proposed knowledge acquisition 

techniques can capture the modelling intentions 

of the user, and how transparent the environment 

can make the process of iterating over the 

modelling task until effective capture is achieved. 

The last of these concerns the issue of how to provide 

useful feedback to the user when the modelling 

process fails to result in a consistent model (due to 

missing, or conflicting, information). Without a consistent 

domain model the plan configuration tools can 

do nothing useful and it is therefore desirable that the 

user be able to develop a correct domain model incrementally. 

We believe this is one of the most interesting 

technical challenges facing the PLANFORM 

project at this stage. 

Bibliography 

[1] AIPS-98 Planning Competition Committee. 

PDDL - The Planning Domain Definition Language. 

Technical Report CVC TR-98-003/DCS 

TR-1165, Yale Center for Computational Vision 

and Control, 1998. 

[2] R. S. Aylett and C. Doniat. The PLANFORM- 

KA Tool. In Proceedings of the 20th UK 

Planning and Scheduling Workshop (PLANSIG- 

2001), Edinburgh, Scotland, 2001. 

[3] R. S. Aylett and C. Doniat. KA Tool and domain 

construction for AI planning applications (submitted 

to ES2002). Technical report, University 

of Salford, 2002. 

[4] R. S. Aylett and C. Doniat. Supporting the Domain 

Expert in Planning Domain Construction 

. In Proceedings of the AIPS’02 Workshop on 

Knowledge Engineering Tools and Techniques 

for AI Planning, 2002. 


[5] R. S. Aylett and S. Jones. Planner and Domain: 

Domain Configuration for a Task Planner. Int. 

Journal of Expert Systems, 9(2):279–318, 1996. 

[6] J. Breuker and B. Wielinga. Models of Expertise 

in Knowledge Acquisition. In Topics in Expert 

Systems Design: Methodologies and Tools 

(eds: G. Guida and C. Tasso), 1989. 

[7] B. Chandrasekaran and T. R. Johnson. Generic 

Tasks and Task Structures: History, Critique 

and New Directions. In Second Generation Expert 

Systems (eds: J. M. David, J. P. Krivine and 

R. Simmons), 1993. 

[8] M. Fox and D. Long. The Automatic Inference 

of State Invariants in TIM. Journal of AI Research, 

9:367–421, 1998. 

[9] M. Fox and D. Long. Hybrid STAN: Identifying 

and Managing Combinatorial Optimization 

Sub-problems in Planning. In Proceedings of 

the International Joint Conference on AI, 2001. 

[10] M. Fox and D. Long. STAN 4: A Hybrid Planning 

Strategy based on Sub-Problem Abstraction. 

AI Magazine, AAAI Press, 22(4), 2001. 

[11] M. Fox and D. Long. Integrating Multiple Subsolving 

Strategies into a Hybrid Planning System 

(submitted to the Journal of AI Research). 

Technical report, Department of Computer Science, 

University of Durham, 2002. 

[12] D. Liu and T. L. McCluskey. The OCL Language 

Manual, Version 1.2. Technical report, 

Department of Computing and Mathematical 

Sciences, University of Huddersfield , 2000. 

[13] D. Long and M. Fox. Extracting Route- 

Planning: First Steps in Automatic Problem Decomposition. 

In AIPS Workshop on Analyzing 

and Exploiting Domain Knowledge, 2000. 

[14] D. Long and M. Fox. Recognising and exploiting 

generic types in planning domains. In Proceedings 

of the International Conference on AI 

Planning and Scheduling, 2000.


[15] D. Long and M. Fox. Recognizing and Exploiting 

Generic Types in Planning Domains). In 

Proceedings of the Fourth International Conference 

on AI Planning and Scheduling (AIPS- 

2000), 2000. 

[16] D. Long and M. Fox. Multi-Processor Scheduling 

Problems in Planning. In Proceedings of the 

International Conference on AI, 2001. 

[17] D. Long, M. Fox, and M. Hamdi. Reformulation 

in Planning. In Proceedings of the Symposium 

on Abstraction, Reformulation and Approximation, 

2002. 

[18] T. L. McCluskey. Planform: Draft Progress Report. 

Technical Report, School of Computing 

and Maths, Univ of Huddersfiled, 2001. 

[19] T. L. McCluskey and D. E. Kitchin. A Tool- 

Supported Approach to Engineering HTN Planning 

Models. In Proceedings of 10th IEEE International 

Conference on Tools with Artificial 

Intelligence, 1998. 

[20] T. L. McCluskey and J. M. Porteous. Engineering 

and Compiling Planning Domain Models to 

Promote Validity and Effici ency. Artificial Intelligence, 

95:1–65, 1997. 

[21] T. L. McCluskey and N. E. Richardson. The induction 

of operator descriptions from examples 

and structural domain knowledge. In Proceedings 

of the 20th Workshop of the UK Planning 

and Scheduling Special Interest Group, pages 

181–192, 2001. 

[22] T. L. McCluskey, N. E. Richardson, and R. M. 

Simpson. An Interactive Method for Inducing 

Operator Descriptions. In The Sixth International 

Conference on Artificial Intelligence 

Planning Systems, 2002. 

[23] L. C. J. Murray. Reuse of Control Knowledge 

in Planning Domains. In AIPS Workshop on 

Knowledge Engineering Tools and Techniques 

for AIP Planning, 2002. 

[24] M. Musen. Modern Architectures for Intelligent 

Systems: Reusable Ontologies and Problem- 

Solving Methods. In AMIA Annual Symposium), 

1998. 

[25] A. Newell. The Knowledge Level. Artificial 

Intelligence, 1982. 

[26] R. M. Simpson, T. L. McCluskey, Maria 

Fox, and Derek Long. Generic Types as 

Design Patterns for Planning Domain Specifications. 

In Proceedings of the AIPS’02 

Workshop on Knowledge Engineering 

Tools and Techniques for AI Planning, 

http://scom.hud.ac.uk/planet/aips02kett/, 2002. 

[27] R. M. Simpson, T. L. McCluskey, D. Liu, and 

D. E. Kitchin. Knowledge Representation in 

Planning: A PDDL to ��¥¡¤£ Translation. In 

Proceedings of the 12th International Symposium 

on Methodologies for Intelligent Systems, 

2000. 

[28] R. M. Simpson, T. L. McCluskey, W. Zhao, 

R. S. Aylett, and C. Doniat. GIPO: An Integrated 

Graphical Tool to support Knowledge 

Engineering in AI Planning. In Proceedings 

of the 6th European Conference on Planning, 

2001. 

[29] M. C. Tully. Object-Oriented Domain Modelling 

for Planning. Technical report, Department 

of Computer Science, University of 

Durham, 2001. 

[30] C. Vogel. Le Genie Cognitif. 1988. 


T.L. McCluskey University of Huddersfield, UK 

Maria Fox University of Durham, UK 

Ruth Aylett University of Salford, UK 

This work was supported by EPSRC projects 

GRM67421/01, GRM70568/01 and GRM68237/01, 

held at the three participating sites.


Second PLANET Gap-Bridging Seminar (GBS-2) 

The second PLANET Gap-Bridging Seminar (GBS- 

2) took place in Delft, The Netherlands, on 21 

November 2002. GBS-2 was held in conjunction 

with PlanSIG2002, the 21st workshop of the 

UK AI Planning & Scheduling Special Interest 

Group. GBS-2 built on the success of the first 

Gap-Bridging Seminar, held in Edinburgh, Scotland, 

in December 2001 (see PLANET News issue 

3, pages 30-31 and http://cswww.essex.ac. 

uk/conferences/planet/GBS-1/). 

The motivation for Gap-Bridging Seminars is the observation 

that industry and academia both work on 

planning and scheduling, but they do not work together 

as well as one would hope. They have different 

goals. Industry must sell, academia must publish, 

and there is no time to talk to each other. PLANET’s 

aim is to give representatives of both realms the opportunity 

and the time to exchange views. 

In the Gap-Bridging Seminars, practitioners from industry 

talk about their work – the techniques, the 


Figure 1: GBS-2 venue in Delft. 

REPORT 

Author: T. Grant 

problems, customer needs, what can and cannot be 

done. This sets research in planning and scheduling 

in a wider context. Most of the speakers stayed 

for the whole of PlanSIG2002, so that attendees 

could have more opportunities to exchange views 

with them. 

In GBS-1 most of the speakers were from companies 

that developed and applied AI-based software products 

for planning and scheduling. By contrast, the 

GBS-2 speakers were chosen to present a wide variety 

of real-world planning and scheduling applications. 

The speakers and their titles were: 

Alessandro Donati European Space Operations 

Centre (ESOC), Darmstadt, Germany: “Space 

Mission Operations Planning and Scheduling: 

Past, present and future”. 

Henk Hesselink and Ron Seljée Dutch Aerospace 

Laboratory (NLR), Amsterdam, Netherlands: “AI 

Planning, Waiting for results?”


Frank Oxener Atos Origin Nederland b.v., Netherlands: 

“Work Force Planning: A logical next step 

after ERP” 

Yossi Rissin and Roman Bartak VisOpt b.v.: 

“When Theory crashed into Reality”. 

In addition, Tim Grant introduced GBS-2 and 

PLANET. 

The audience consisted of 42 participants, mostly 

from the academic community (see Figure 1). UK 

and the Netherlands each provided a third of the 

people attending. The remaining third was divided 

over Spain, Italy, Germany, France, Belgium, and the 

Czech Republic. Academic attendees were primarily 

post-graduate students. 

Alessandro Donati identified the need for three communities 

– users, innovators and implementers – to 

work more closely with one another. He explained 

that ESOC is the part of the European Space Agency 

(ESA) responsible for launching and operating scientific 

spacecraft. A very recent survey at ESOC had 

shown that the planning and scheduling processes 

differed radically between space missions. The reasons 

why had yet to be established. A priori, it would 

Figure 2: Air traffic management problem in Europe. 

seem to be more efficient to have a standard process 

and standard tools for all missions. He welcomed the 

involvement of the research community (the innovators) 

and PLANET, and mentioned the opportunities 

for post-graduate students to work in ESA. 

A lively discussion ensued, with researchers calling 

for information to be made available on complex, 

real-world planning and scheduling domains 

and problems as case studies. This discussion resulted 

in a lunchtime splinter meeting on the second 

day between Alessandro and representatives of 

the innovator and implementer communities on how 

PLANET and ESOC could co-operate to document 

one or more case studies, to mutual benefit. 

Henk Hesselink talked about NLR’s experience in 

providing software support for planning and scheduling 

in civil and military aircraft operations. The 

NLR is a non-profit organisation founded in 1919 

to provide technical and scientific contributions to 

aerospace organisations in the private and public sectors. 

Turnover is EUR 70 million annually, split 65% 

civil and 35% military. The ratio between development 

and operations is 40:60, and between aeronautical 

and space applications is 85:15. Facilities


available to the NLR include large wind-tunnels, two 

aircraft, various simulators, and computing environments 

(including a supercomputer). NLR co-operates 

with Amsterdam Schiphol airport, Dutch industry, 

and Eurocontrol. 

Henk focussed on the forthcoming air traffic management 

(ATM) problem in Europe (see Figure 2). This 

problem is not recognised by the controllers – the potential 

end-users – because ATM is divided into subproblems. 

Many actors are involved, and each airport 

is different, yet has features in common. Currently, 

ATM controllers do not plan, but work on a 

“first-heard, first-served” basis. They are conservative, 

with air safety being uppermost in their mind, 

followed by the need for fairness in allocating resources. 

They have neither the time nor the interest 

to talk to researchers. Henk argued that gap-bridging 

had to start with the foundations of the bridge: finance 

and user interest. He advocated the use of prototypes 

to make users aware of their problems and 

potential solutions. 

Frank Oxener explained how the complexity of recent 

law-making in the Netherlands had given rise to 

commercial opportunities for planning and scheduling 

applications in work-force management (see Figure 

3). Several Dutch companies were offering software 

products and associated services. 

One law required employers to give shift-workers 36 

hours of rest in any 7-day period or 60 hours rest in 

any 9-day period. Once every 5 weeks they had to 

be allowed at least 32 hours continuous rest. Another 

law provided for irregularity payments of 20% 

of salary to workers working Monday to Friday between 

06:00 and 08:00, providing they started before 

07:00. On Saturdays the payment was 40%, and on 

Sundays 100%. These laws were a challenge to software 

systems for recording actual hours worked and 

planning rest periods and irregularity payments. He 

outlined his experiences with applying a variety of 

commercially-available software packages. 

Frank elicited audience participation by asking his 

listeners to write down a question on a sheet of paper. 

He then answered each question. Sample questions 

included: 


Can the tools easily be adapted to the workinghour 

laws in other countries? 

What is the key feature in work-planning software 

for user acceptability? 

What type of techniques are used to solve the 

problems you have described? 

How does a different culture really affect a planning 

process? 

How might a planning solution be overridden by 

labour union objections? 

How long does it take to implement a system from 

first requirement to operational use? 

Are your solutions sufficiently flexible to deal 

quickly with changes in the law? 

What is coming after workflow management? 

Why? 

Roman Bartak focussed on the human factors issues. 

Human behaviour is inconsistent and readily affected 

by mood, environment and psychological pressure. It 

can only be modelled statistically. Plant personnel 

and planners are motivated by pride, their position in 

the organisation, and future job security. Pride makes 

it difficult for users to admit mistakes, problems and 

weaknesses. They protect their position by being nice 

to superiors, gaining professional respect, and trying 

to serve many masters at the same time. They protect 

their job by withholding knowledge. Internal politics 

and power-plays are key factors in decision-making. 

Figure 3: Right people, right place? 

(acknowledgements to Parallax b.v.)


These factors and issues make academic research irrelevant. 

One scheduling expert told him: “I have 

never seen a Job Shop Scheduling problem in practice”. 

From an academic point of view, the ideal factory 

is one that is totally automated, populated with 

robots and Automated Guided Vehicles (AGVs). Alternatively, 

it might be a new factory that has not yet 

been put into operation. There is then no previous 

“know-how”, rules and procedures, bad habits, and 

day-to-day reality to confront theory. 

Roman contrasted the views of planners and academics, 

looking at the “Not Invented Here” syndrome. 

Summarising the lessons he had learned at 

VisOpt b.v., he advocated the development of a visual 

modelling language as a way of improving communication 

between the two communities (see Figure 4). 

PLANET provided sponsorship for 14 post-graduate 

students from Spain, Germany, Netherlands, UK, 

Greece, and Italy to attend GBS-2. In return, the students 

wrote a short report on what they had learned 

from GBS-2 and PlanSIG2002. Key quotes included: 

“The gap turned out to be bigger than I thought”. 

Figure 4: VisOpt’s visual modelling language. 

“The workshop gave me the chance to get acquainted 

to people whose work I had been reading 

in papers for quite some time”. 

“Industry does not always need the best possible 

plan – it needs one that is good enough”. 

“An attempt should be made from both sides to 

approach one another”. 

“Managers and planners need to be convinced to 

change to new planning methods”. 

“Industry (NASA, ESOC, NLR, etc) can help 

academia by supplying real planning domains, 

problems, and case studies”. 

“[There are] four key items: visual modelling, 

planning as a step-by-step process, scheduling as 

a process to reason on resource constraints, and 

the need for generating good enough plans in a 

reasonable time”. 

“There should be cooperation between [human] 

planner and computer”.


“[It is] necessary for researchers to show the benefits 

that can be obtained”. 

“There is also a gap between different academic 

communities: Operations Research and AI”. 

“Employees do not see the global view, but only 

the bit they are working on”. 

“One challenging line of research is the validation 

of planning domains”. 

“Domains in industry are far less predictable 

and much more dynamic that those used by researchers”. 

“Both communities need more communication”. 

“[GBS-2] encouraged new and interesting discussions, 

thus exposing many unexplored areas of research, 

and providing good ideas for future work”. 


“People from the two sides have to come closer 

and cooperate since this will be of great profit for 

both”. 

In summary, both speakers and audience came away 

from the second PLANET Gap-Bridging Seminar 

having learned more about the academic and industrial 

aspects of planning and scheduling. All attendees 

were most grateful to PLANET and the CEC for 

making it possible to exchange such a diversity of 

views, and look forward with eagerness to the third 

Gap-Bridging Seminar. 


Tim Grant Atos-Origin Nederland b.v., 

Nieuwegein, The Netherlands, Tim.Grant@ 

atosorigin.com


ICAPS 2003 

ANNOUNCEMENT


3rd PLANET International Summer School 2003 

The International Summer School on AI Planning is 

a great opportunity for Ph.D. students and young researchers 

to be exposed to introductory and advanced 

courses on various aspects of Artificial Intelligence 

Planning, and to spend time and discuss research directions 

with their colleagues and with the teachers, 

foremost researchers in the field. 

The first edition of the school was held in Cyprus in 

September 2000, while the second school was held 

in Halkidiki, Greece, in September 2002. The third 

edition of the school will be held on the mountains 

of Trentino, in the northern part of Italy, in June 

2003. The school will be colocated with International 

Conference on Automated Planning and Scheduling 

(ICAPS’03). 

The Program Chairs of the school are Daniel Borrajo 

ICAPS 2003 – Doctoral Consortium 

ICAPS-2003 invites PhD students to apply for the 

Doctoral Consortium, which will provide an opportunity 

for a group of students to discuss and explore 

their research interests and career objectives with established 

researchers in Planning and Scheduling. 

The aims of the Doctoral Consortium are the following: 

to provide a forum for students to present their 

current research, and receive feedback from other 

students and senior researchers; 

to promote contacts among PhD students working 

in the same area; 

to support students with information and advice 

on academic, research and industrial careers; 

to financially support students by covering the 

conference registration fee and by partially contributing 

to travel expenses. 


ANNOUNCEMENT 

Millan (Universidad Carlos III de Madrid, Spain), 

and Alessandro Cimatti (ITC-irst, Italy). ITC-irst 

is also responsible for the local organization of the 

event, with a team composed by Piergiorgio Bertoli, 

Mark Carman, Alessandro Cimatti and Alessandro 

Tuccio. 

Additional, up-to-date information will be available 

at the official web site of the school: 

http: 

//sra.itc.it/planet/summer-school-03/ 

Programme 

CALL FOR PARTICIPATION 

The programme will consist of students’ presentations 

on their current research interests. A voluntary 

mentoring programme will be organized to link students 

with like-minded researchers. 

Submissions 

We encourage submissions from Ph.D. students at 

any level, and from any topic area and methodology 

within Planning and Scheduling. On the basis of 

the submissions, the Organizing Committee will select 

a group of students that will be invited to present 

their work during the Doctoral Consortium, and also 

to present a poster at the ICAPS-2003 poster session. 

Students accepted for participation in the Doc


toral Consortium will have free conference registration 

and a fixed allowance for travel/housing. The 

students’ abstracts will be made available on the web 

and included as part of the conference proceedings. 

Applicants should submit an extended abstract of 

5 pages maximum by email to one of the Doctoral 

Consortium chairs. The submission should be 

in AAAI style format (http://www.aaai.org/ 

Publications/Author/macros-link.html) 

and sent either as a PostScript or as a PDF file. It 

should describe original, unpublished work, completed 

or in progress, that is part of the doctoral work 

of the student. If an extended version of the paper 

is also submitted to the technical programme, please 

indicate it in the submission. Double submission is 

acceptable, but if the paper is accepted for the technical 

programme, the student will present the work 

only in the technical programme sessions and not 

during the Doctoral Consortium. 

In addition, the dissertation advisor should send a 

letter of recommendation by e-mail to one of the 

Doctoral Consortium chairs. It should include the 

expected date for thesis submission, and the motivation/expected 

benefit for the student to attend the 

ICAPS 2003 – Workshop Program 

The ICAPS-03 workshop program (June 9-10, before 

the main program), has been set. There will 

be five workshops, covering a broad range of topics 

ranging from the current and future state of the 

Planning Competition, to issues arising as planning 

and scheduling are applied to ever-more-complex domains, 

to specific application areas. 

The ICAPS-03 workshops: 

Planning and Web Services 

Web services are revolutionizing the way industry 

and government operate. Web services both pro- 

Doctoral Consortium. This letter can be sent in as 

either a text or a PostScript or a PDF file. 

Important Dates 

March 31st: deadline for submitting papers and 

letters of support 

April 18th: notification of acceptance to program 

April 25th: camera ready copy of the papers to the 

chairs 

Doctoral Programme Chairs 

Jeremy Frank , NASA Ames Research Center, 

frank@email.arc.nasa.gov 

Susanne Biundo , University of Ulm, susanne. 


Additional information is available at the ICAPS 

Website 

http://icaps03.itc.it 

This event will be sponsored by PLANET and Nasa. 

ANNOUNCEMENT 

vide information (e.g., available flights) and change 

the world (e.g., buying a flight ticket). As the 

Web evolves into the Semantic Web, the myriad of 

available services are being described declaratively. 

Machine-understandable descriptions enable the automatic 

discovery, use, and composition of web services. 

With increased interest in the web services paradigm, 

composition of web services has become of primary 

importance. Several languages for describing web 

services and their composition are currently being defined 

and seek to become standards. From a planning 

perspective, the web services can be seen as op-


erators, specific web services compositions as plans, 

and automatic web service composition as a form of 

planning. This workshop will provide planning researchers 

with a forum for presenting planning results 

relevant to web services, identify new challenges, and 

lead the development of the critically important field 

of web services. 

Jose Luis Ambite 

Workshop on Plan Execution 

Much work in the planning community has focussed 

primarily upon developing efficient ways of generating 

plans that are not actually executed. This 

was reflected in the AIPS 2002 Planning Competition 

which measured the efficiency and optimality of 

plans that were generated, but not executed. As execution 

may not result in the intended outcome the 

sequence of actions that is eventually executed may 

not be as valuable as that in the original plan, which 

leads to the question as to whether it is necessary to 

generate plans that are near optimal. Researchers in 

the planning community are increasingly concerned 

with executing plans and the design of systems in 

which planning and execution are continually interleaved 

and actively managed. 

This workshop is intended for researchers who have 

interests in plan execution in a variety of domains 

such as robotics, space applications, information 

gathering and other areas. 

Alex Coddington 

Workshop on PDDL 

PDDL, originally developed by Drew McDermott 

and the 1998 planning competition committee, was 

inspired by the need to encourage the empirical comparison 

of planning systems and the exchange of 

planning benchmarks within the community. Its development 

improved the communication of research 

results and triggered an explosion in performance, 

expressivity and robustness of planning systems. 

PDDL has become a de facto standard language for 

describing planning domains, not only for the com- 


petition but more widely, as it offers an opportunity 

to carry out empirical evaluation of planning systems 

on a growing collection of generally adopted standard 

benchmark domains. The emergence of a language 

standard will have an impact on the entire field, influencing 

what is seen as central and what peripheral 

in the development of planning systems. The adoption 

of PDDL in this role is itself an issue for debate: 

perhaps a completely different modelling language 

is called for. We believe that it is therefore important 

to provide a forum in which the community can 

give feedback and present their ideas to the language 

designers, and in which the language designers can 

discuss their ideas for maintaining and extending, or 

even replacing the language. 

Sylvie Thiebeaux 

Planning under Uncertainty and Incomplete 

Information 

Controlling intelligent agents in complex real-world 

environments poses requirements that are not addressed 

in classical AI planning. Often it is not sufficient 

to find a sequence of actions leading to a given 

goal, since the initial state may not be known with 

precision, and action effects cannot be predicted with 

certainty. 

In the past few years, there has been a growing interest 

in more general planning techniques, able to 

tackle the problems of uncertainty, nondeterminism, 

and incompleteness of information. Several research 

works have proposed more expressive domain models 

and description languages (e.g., allowing for actions 

with multiple transitions, possibly with different 

probabilities, and with costs), and more complex 

models of execution (e.g., dealing with information 

gathering at run-time). New planning techniques and 

algorithms have been developed to operate on such 

extended models and to produce plans which achieve 

the goals despite the uncertainty and incompleteness 

of information. The goal of this workshop is to bring 

together people working in different areas of planning 

under uncertain and incomplete information. 

Marco Pistore


The Planning Competition: Impact, 

Organization, Evaluation, Benchmarks 

The planning competition series undoubtedly has had 

a huge impact on the field of AI planning, including 

such aspects as growing standardization of complex 

planning domain description languages, dramatically 

improved scalability of existing approaches, and a 

growing database of commonly used benchmark examples. 

It is therefore important to provide an opportunity 

for discussing topics related to the competition. 

The workshop aims at doing just this. We want 

to collect together panels on topics such as the role of 

the competition in and for the field, organizational as- 

ICAPS 2003 – Tutorials 

Provisional Schedule 

Monday, June 9 

pects of the competition, (competition) results evaluation, 

and benchmarking issues. Technical presentations 

are planned on topics related to the competition 

such as language alternatives, and methods of empirical 

evaluation. 

Joerg Hoffmann and Stefan Edelkamp 

Submissions to all workshops are due by 31 March, 

2003, with notification and submission of camaraready 

versions by the end of April. 

Additional detail about the workshops, as well as further 

information about ICAPS-03 in general, is also 

available on the conference website: 

http://icaps03.itc.it/ 

morning afternoon 

Timed Automata for Planning and Scheduling 

Oded Maler (Verimag) 

Tuesday, June 10 

morning afternoon 

Practical Approaches to Handling Uncertainty in 

Planning and Scheduling 

J. Christopher Beck (University College Cork) 

and Thierry Vidal (ENIT) 

Model Checking – A Hands-On Introduction 

Alessandro Cimatti, Marco Pistore and Marco 

Roveri (ITC- IRST) 

ANNOUNCEMENT 

Resource-Bounded and Time-Critical Reasoning 

Lloyd Greenwald (Drexel University) and 

Shlomo Zilberstein (University of Massahusetts) 

ICAPS’03 Tutorial chairs: Anthony Barrett, NASA Jet Propulsion Laboratory 

Jussi Rintanen, Albert-Ludwigs-Universität Freiburg


Timed Automata for Planning and 

Scheduling 

In this tutorial we propose the model of timed automata, 

originating from the verification of real-time 

systems, as a model for posing and solving timedependent 

planning and scheduling problems. We 

believe that in the same sense as automata are used 

as the major vehicle for verification of systems where 

the model of time is qualitative, timed automata can 

be the center of a a unifying mathematical modeling 

framework for quantitative time, having the following 

attractive features: 

1. It is sufficiently expressive to describe the essential 

aspects of time-dependent real-life problems 

in a variety of application domains. 

2. It provides for models with well-defined and clear 

dynamic semantics. 

3. These models are amenable to computer-aided design 

methods such as simulation, testing, verification 

and automatic synthesis of (optimal) schedules 

and plans. 

4. These methods are currently supported by tools 

of various levels of maturity, that treat the specific 

computational problems of time-related reasoning. 

Oded Maler was born in 1957 in Haifa, Israel. He 

obtained his B.A. in Computer Science from the 

Technion, Haifa in 1979 and his M.Sc. in Management 

Science from the University of Tel-Aviv at 

1984. In 1989 he finished his Ph.D. thesis (Finite 

Automata: Infinite Behavior, Learnability and Decomposition 

), under the supervision of A. Pnueli in 

the department of Applied Mathematics and Computer 

Science, Weizmann Institute, Rehovot. After 

two years of post-doc at IRISA, Rennes, he moved 

to Grenoble at 1992 and obtained a research position 

(CR1) at the CNRS (French National Center of Scientific 

Research) in 1994. He has been promoted to 

“research director” (DR2) in 2001. Dr. Maler’s research 

is centered around the theory of automata and 


its various extensions, most notably timed automata, 

hybrid automata and their application to control, embedded 

systems, scheduling and circuit timing analysis. 

Practical Approaches to Handling 

Uncertainty in Planning and Scheduling 

This tutorial presents techniques for dealing with the 

fact that the execution of plans and schedules in the 

real world cannot assume a static environment: the 

world changes non-deterministically during problem 

solving and execution. We present techniques from 

the Artificial Intelligence and Operations Research 

literature for handling uncertainty in planning and 

scheduling with emphasis on practical techniques. 

Such techniques include reactive, on-line scheduling 

and planning, and proactive, off-line techniques that 

build solutions that can cope with uncertain events, 

as well as intermediate approaches between these extremes. 

J. Christopher Beck received a PhD in Artificial Intelligence 

in 1999 from the University of Toronto under 

the supervision of Mark S. Fox. From 1994 to 

1999 he was the project manager of the Intelligent 

Scheduling Research Group at the Enterprise Integration 

Laboratory at University of Toronto. The 

focus of his research was measurements of problem 

structure as a basis for scheduling heuristics within a 

constraint-based scheduling framework. From 1999 

until 2002 he was a software developer and Senior 

Scientist on the Scheduler team at ILOG, SA in Gentilly, 

France. As of June, 2002, he moved to the position 

of Staff Scientist at the Cork Constraint Computation 

Centre, University College Cork. His research 

interests focus on problem structure, hybrid 

algorithms, search in constraint-directed scheduling, 

and in the extension of constraint modeling and solving 

capabilities to incorporate aspects of real-world 

scheduling such as uncertainty, dynamic arrival of activities, 

and robustness. 

Thierry Vidal received a PhD in Artificial Intelligence 

in 1995 from the University of Toulouse under 

the supervision of Malik Ghallab. He had worked


in the Robotics and AI team of the LAAS-CNRS 

in Toulouse, France, working on temporal constraint 

processing in temporal planning (the IxTeT system) 

and in task scheduling, with a special focus on uncertain 

durations. In 1996-97 he was a guest researcher 

in Erik Sandewall’s team at the Department 

of Computer Science of the University of Linköping, 

Sweden, where he conducted basic research work in 

the area of on-line decision making through contingent 

plan recognition and reactive controller synthesis. 

From 1997 he is assistant professor at ENIT 

in Tarbes, France, working in the Automated Production 

team of the Production Engineering Laboratory, 

with external collaborations with Hélène Fargier 

(Possibilistic Reasoning team, IRIT, Toulouse), Paul 

Morris (NASA Ames Research Center, California, 

USA), and Ioannis Tsamardinos and Martha Pollack 

(University of Pittsburgh, USA). His current research 

interests are uncertain constraint reasoning in 

planning, scheduling and resource allocation, multiagent 

approaches to scheduling, reactivity, conditional 

planning and robust scheduling. 

Resource-Bounded and Time-Critical 

Reasoning 

A central problem in artificial intelligence is how to 

develop computational models that allow decisionsupport 

systems or autonomous agents to react to a 

situation after performing the right amount of deliberation. 

Frequently, the complexity of problem 

solving makes it beneficial to use approximate solutions 

rather than try to compute the optimal answer. 

This issue arises in a wide range of application 

domains including medical trauma management, 

Bayesian inference, sequence alignment, graphics 

rendering, web page prefetching, autonomous space 

exploration, real-time avionics, and robot navigation. 

This tutorial explores the theory and practice of building 

intelligent systems that reason explicitly about 

employing limited computational resources to generate 

timely solutions to difficult combinatorial optimization, 

planning and scheduling problems. Solution 

techniques go beyond simple greedy or reactive 

algorithms to achieve high-quality solutions while 

meeting both hard and soft real-time deadlines. We 

will explore over fifteen years of progress in this area, 

covering historical perspectives, state-of-the-art solution 

techniques, and current and future challenges. 

Participants should be familiar with introductory artificial 

intelligence, algorithm design and analysis, and 

introductory probability and statistics. 

Lloyd Greenwald is an Assistant Professor of Computer 

Science and Director of the Intelligent Time- 

Critical Systems Lab at Drexel University. He received 

his Ph.D. in Computer Science from Brown 

University. His research interests include timecritical 

planning and scheduling, mobile robotics, 

machine learning, ad hoc and sensor networks, and 

medical decision making. 

Shlomo Zilberstein is an Associate Professor of 

Computer Science and Director of the Resource- 

Bounded Reasoning Lab (http://anytime. 

cs.umass.edu) at the University of Massachusetts, 

Amherst. He received his Ph.D. in Computer 

Science from the University of California, 

Berkeley. His research interests include approximate 

reasoning, decision theory, heuristic search, planning 

and scheduling, and resource-bounded reasoning. 

Model Checking – A Hands-On 

Introduction 

Model Checking is a formal technique for the verification 

of designs of concurrent systems. It is based 

on the representation of the system being analyzed as 

a (finite state) transition systems (e.g. Kripke models), 

while the requirements are typically expressed 

in temporal logics. A system satisfies a given property 

amounts to checking if the corresponding temporal 

formula is true in the Kripke model. Model 

checking is very effective in pinpointing design errors 

that are extremely hard to detect by means of testing, 

and is therefore being applied in the industrial 

development of reactive systems, hardware designs, 

and communication protocols. Furthermore, model 

checking techniques and tools are gaining interest


in several fields of Artificial Intelligence (e.g. Planning, 

Multi-agent systems, and Model-based Diagnosis) 

and Engineering (e.g. Requirement Verification, 

Safety Analysis). Of particular interest is Symbolic 

Model Checking, which makes it is possible to analyze 

extremely large finite-state systems by means of 

symbolic representation techniques (e.g. Binary Decision 

Diagrams, propositional satisfiability). 

Alessandro Cimatti is the leader of the formal methods 

group within the Automated Reasoning Systems 

division (SRA) at ITC-IRST (http://www. 

irst.itc.it). The activities carried out by the 

group include basic research, the development of 

the NuSMV (http://nusmv.irst.itc.it) 

model checker, and technology transfer in industrial 

projects in the areas of safety critical applications 

(e.g., railways, avionics, aerospace, industrial plant 

controllers). 

Alessandro Cimatti has participated in and led several 

industrial projects aimed at the use of formal methods 

for the development and verification of safety critical 

systems and embedded controllers. Some examples 

are the validation of Interlocking Systems, the development 

of Rail Traffic Management Systems, the design 

of tools for on-board verification, and the verification 

of safety-critical communication protocols. 

Alessandro Cimatti is the leader of the development 

of NuSMV. His main research interests include 

the development of advanced model checking techniques, 

and the application of model checking for the 

synthesis of reactive controllers and test cases. He 

has also contributed to the research in theorem proving, 

formal languages for the specification of multiagent 

systems, planning and robotics. 

Marco Pistore is Associate Professor at the Department 

of Information and Communication Technolo- 


gies of the University of Trento (http://www. 

dit.unitn.it/) and Research Consultant at 

ITC-IRST (http://www.irst.itc.it). His 

research interests are in formal methods and in the 

application of formal methods to planning and to synthesis 

of controllers. Marco Pistore has been the responsible 

of the development of the NuSMV checker. 

He is also working to the Formal Tropos project, aiming 

at the development of a formal language and of 

formal analysis techniques for the verification of requirements 

specifications. Marco Pistore has also 

participated to research and industrial projects on the 

application of formal methods to the design and verification 

of safety-critical systems and of embedded 

controllers. 

Marco Roveri received a PhD in Computer Science 

in 2002 from the University of Milano in collaboration 

with ITC-Irst under the supervision of 

A. Cimatti. His PhD thesis “Planning in Non- 

Deterministic Domains via Symbolic Model Checking” 

was awarded by the Italian Association for Artificial 

Intelligence (AI*IA) the best price for PhD 

thesis in artificial intelligence in Italy. He his in the 

steering committee of the NuSMV symbolic model 

checker, the first state-of-the-art open source symbolic 

model checker. Since 2001 he is working at 

ITC-Irst in the Automated Reasoning Systems division. 

From 1997 until 2002 he was collaborating 

with ITC-Irst on topics related to Artificial Intelligence 

Planning and Formal Verification. His research 

interest are: integration of formal verification techniques 

along the whole software development process, 

planning in non-deterministic domains under 

different assumptions on run-time observability using 

symbolic model checking techniques and symbolic 

model checking.


ANNOUNCEMENT 

Invitation to Participate in TAC’03 - A Supply Chain Trading 

Competition 

Authors: R. Arunachalam, N. Sadeh, E. Aurell, J. Eriksson, N. Finne, and S. Janson 

Supply chain management is concerned with planning 

and coordinating the activities of organizations 

across the supply chain, from raw material procurement 

to finished goods delivery. In today’s global 

economy, effective supply chain management is vital 

to the competitiveness of manufacturing enterprises 

as it directly impacts their ability to meet changing 

market demands in a timely and cost effective manner. 

With annual worldwide supply chain transactions 

in the trillions of dollars, the potential impact 

of performance improvements is tremendous. While 

today’s supply chains are essentially static, relying 

on long-term relationships among key trading partners, 

more flexible and dynamic practices offer the 

prospect of better matches between suppliers and customers 

as market conditions change. Adoption of 

such practices has however proven elusive, due to the 

complexity of many supply chain relationships and 

the difficulty in effectively supporting more dynamic 

trading practices. TAC-03 was designed to capture 

many of the challenges involved in supporting dynamic 

supply chain practices, while keeping the rules 

of the game simple enough to entice a large number 

of competitors to submit entries. The game has been 

designed jointly by a team of researchers from the 

e-Supply Chain Management Lab at Carnegie Mellon 

University and the Swedish Institute of Computer 

Science (SICS). 

Specifically, TAC-03 features rounds where eight PC 

assembly agents compete for customer orders and for 

procurement of a variety of components. Customers 

issue requests for quotes and select from quotes submitted 

by the PC assemblers, based on delivery dates 

and prices. The assembly agents are limited by the 

capacity of their assembly lines and have to procure 

components from a set of possible suppliers. The 

game distinguishes between four types of components: 

CPUs, Motherboards, Memory Units and Disk 

drives. It features a variety of components of each 

type (e.g. different CPUs, different motherboards, 

etc.). Customer demand comes in the form of requests 

for quotes for different types of PCs, each requiring 

a different combination of components. 

The PC assembly agents compete over a relatively 

long period of time during which customer demand 

and availability of supplies varies according to predefined 

stochastic distributions. The aim of each competitor 

agent (PC assembly agent) is to maximize its 

profit, by (1) competing with other agents for valuable 

customer orders and profitable supplier commitments, 

and (2) managing the assembly of products to 

meet its existing customer delivery commitments. 

The game is representative of a broad range of supply 

chain situations. It is challenging in that it requires 

agents to concurrently compete in multiple markets 

(markets for different components on the supply side 

and markets for different products on the customer 

side) with interdependencies and incomplete information. 

It allows agents to strategize (e.g. specializing 

in particular types of products, stocking up components 

that are in low supply). To succeed, agents 

will have to demonstrate their ability to react to variations 

in customer demand and availability of sup-


plies, as well as adapt to the strategies adopted by 

other competing agents. 

We would like to invite you to consider submitting 

an entry to the competition. This is a unique opportunity 

to develop and evaluate supply chain trading 

technology in a competitive environment. Entrants 

will also be invited to submit articles in an upcoming 

book and will benefit from the publicity associated 

with the event in the form of press coverage and the 

publication of an AI magazine article discussing the 

competition. 

A detailed game description, including the rules 

of TAC03, will be published on the TAC website 

(http://www.sics.se/tac/) by late November 

2003. The TAC 03 game server and some simple PC 

assembly agents will be available for practice games 

on the TAC website by February 1, 2003. This will 

enable prospective agent designers to test and finetune 

their designs by playing practice games. The 

competition itself will be played in a format similar 

to earlier TAC games with eight agents competing 

in each round. Qualification rounds will be 

held in May 2003 with the finals slated to take place 

at IJCAI-03 in Acapulco in August. Stay tuned on 

(http://www.sics.se/tac/) for more information. 

Raghu Arunachalam and Norman Sadeh 

(CMU), 

Erik Aurell, Joakim Eriksson, Niclas Finne, 

and Sverker Janson (SICS) 

JOB OPENING 

Postdoctoral and Doctoral Positions at Australian National ICT 

Center 

The Australian National ICT Center (NICTA) is a 

new research institute jointly set up by the University 

of New South Wales (UNSW) and the Australian 

National University (ANU) with respective nodes in 

Sydney and Canberra. It is funded by the Australian 

Federal and State Governments, in partnership 

with the two universities and industry. NICTA will 

host top-ranked international researchers and graduate 

programs, and will cover major areas in computing, 

systems and telecommunications. 

The NICTA home page is 

http://www.nicta.com.au 

In it there are links to the existing programs (more 

will be added in the future), and vacant positions. 

The activities of the Knowledge Representation 

and Reasoning (KRR) program (http://www. 

nicta.com.au/kr.html) are typified by the 


content of papers appearing in, say, the KRR section 

of the IJCAI proceedings, with an additional focus on 

planning and constraints. The program has the preexisting 

multi-university Knowledge Systems Group 

(KSG) as its initial core, but now seeks to expand by 

recruiting research personnel and graduate doctoral 

students. At the moment, post-doctoral fellows and 

doctoral students are sought: for information on how 

to apply and conditions please follow the Positions 

Vacant link in the NICTA home page. 

KRR welcomes preliminary inquiries about other 

levels of personnel. 

Norman Foo, Maurice Pagnucco (UNSW), Sylvie 

Thiebaux (ANU) 

Dr. Sylvie Thiebaux Research Fellow, RSISE, 

The Australian National University, Canberra ACT 

0200, Australia 

http://csl.anu.edu.au/˜thiebaux 

sylvie.thiebaux@anu.edu.au 

Tel: +61 (2) 6125 8678, Fax: +61 (2) 6125 8651


Member List for PLANET 

Currently, PLANET has 58 nodes from 15 European 

countries. Sites and contact persons are: 

Austria 

- XIMES GmbH, Johannes Gärtner, 

gaertner@ximes.com 

Belgium 

- Robonetics NV, Filip Verhaeghe, 

filip.verhaeghe@roboentics.com 

- Space Applications Services (SAS), Richard 

Aked, 

ra@sas.be 

Cyprus 

- University of Cyprus, Yannis Dimopoulos, 

yannis@cs.ucy.ac.cy 

Czech Republic 

- Charles University, Praha, Roman Barták, 

bartak@kti.mff.cuni.cz 

France 

- COSYTEC S.A., Abderrahmane Aggoun, 

abderrahmane.aggoun@cosytec.com 

- ILOG S.A., Philippe Laborie, 

laborie@ilog.fr 

- Laboratoire d’ Analyse et d’ Architecture des Systemes 

(LAAS-CNRS), TCU Robot Planning, Malik 

Ghallab, 

malik@laas.fr 

- Laboratoire d’ Informatique Marseille (LIM- 

CNRS), Camilla Schwind, 

schwind@lim.univ-mrs.fr 

- MASA Group, Emmanuel Chiva, 

emmanuel.chiva@masagroup.net 

- ONERA Systems Control and Flight Dynamics 

Department, TCU On-line Planning and 

Scheduling, Gérard Verfaillie, 

Gerard.Verfaillie@cert.fr 

INFORMATION 

- THOMSON-CSF, Simon De Givry, 

simon.degivry@thalesgroup.com 

Germany 

- University of Ulm, Coordinating Node, Susanne 

Biundo, 


- Aachen University of Technology, Gerhard Lakemeyer, 

gerhard@cs.rwth-aachen.de 

- University of Bonn, Armin Cremers, 

abc@informatik.uni-bonn.de 

- Bremer Institut für Betriebstechnik und angewandte 

Arbeitswissenschaft (BIBA), Frithjof Weber, 

web@biba.uni-bremen.de 

- Darmstadt University of Technology, Ulrich 

Scholz, 

scholz@informatik.tu-darmstadt.de 

- German Research Center for Artificial Intelligence 

(DFKI), Markus Meyer, 

meyer@dfki.de 

- University of Freiburg, Bernhard Nebel, 

nebel@informatik.uni-freiburg.de 

- Fraunhofer - Autonomous intelligent Systems 

(AiS), Joachim Hertzberg, 

hertzberg@ais.fraunhofer.de


- Technical University of Munich, Michael Beetz, 

Michael.Beetz@informatik.tu-muenchen. 

de 

- Siemens AG, Wendelin Feiten, 

wendelin.feiten@mchp.siemens.de 

Greece 

- Aristotle University of Thessaloniki, Ioannis Refanidis, 

yrefanid@csd.auth.gr 

- Foundation for Research and Technology - Hellas 

(ICS-FORTH), Dimitrios Plexousakis, 

dp@csi.forth.gr 

- National Centre for Scientific Research 

”Demokritos”, Constantine Spyropoulos, 

costass@iit.demokritos.gr 

- Technical University of Athens (ICCS), Spyros 

Tzafestas, 

tzafesta@softlab.ece.ntua.gr 

- Technical University of Crete, Manolis 

Koubarakis, 

manolis@ced.tuc.gr 

- University of Ioannina, Chrysostomos Stylios 

stylios@cs.uoi.gr 

Hungary 

- Computer and Automation Research Institute 

Hungarian Academy of Sciences (MTA SZTAKI), 

László Monostori, 

laszlo.monostori@sztaki.hu 

Italy 

- DEIS - University of Bologna, Paola Mello, 

pmello@deis.unibo.it 

- University of Brescia, Alfonso Gerevini, 

gerevini@ing.unibs.it 

- DIST - University of Genoa, Enrico Giunchiglia, 

Enrico@dist.unige.it 

- Consiglio Nazionale delle Ricerche - Istituto di 

Psicologia (IP-CNR), TCU Aerospace Applications, 

Amedeo Cesta, 

cesta@ip.rm.cnr.it 


- University of Perugia, TCU Planning & Scheduling 

for the Web, Alfredo Milani, 

milani@dipmat.unipg.it 

- Istituto per la Ricerca Scientifica e Tecnologia 

(IRST), Paolo Traverso, 

traverso@irst.itc.it 

- University of Parma, Agostino Poggi, 

poggi@ce.unipr.it 

The Netherlands 

- Delft University of Technology, Cees Witteveen, 

witt@cs.tudelft.nl 

- NLR – National Aerospace Laboratory, Henk Hesselink, 

hessel@nlr.nl 

Portugal 

- Instituto Superior de Engenharia do Porto 

ISEP/IPP, João Rocha, 

jrocha@ipp.pt 

SLOVENIA 

- University of Maribor, Peter Kokol, 

kokol@uni-mb.si 

Spain 

- iSOCO, Intelligent Software for the Networked 

Economy, Antonio Reyes Moro, 

toni@isoco.com 

- Technical University of Catalonia, Lluís Vila, 

vila@lsi.upc.es 

- University of Granada, Luis Castillo, 

L.Castillo@decsai.ugr.es 

- University Carlos III of Madrid, TCU Workflow 

Management, Daniel Borrajo, 

dborrajo@ia.uc3m.es 

- Universitat Politècnica de Catalunya, Institut de 

Robòtica i Informàtica Industrial, Tom Creemers, 

creemers@iri.upc.es 

- Universidad Politecnica de Valencia, Eva Onaindia, 

onaindia@dsic.upv.es


- Universitat Rovira i Virgili, Tarragona, Miguel 

Angel Garcia, 

magarcia@etse.urv.es 

Sweden 

- Linköping University, Patrick Doherty, 

patdo@ida.liu.se 

- Örebro University, Alessandro Saffiotti, 

alessandro.saffiotti@aass.oru.se 

United Kingdom 

- British Telecommunications, David Lesaint, 

david.lesaint@bte.bt.com 

- University of Durham, Julie Porteous, 

j.m.porteous@durham.ac.uk 

- University of Essex, Sam Steel, 

sam@essex.ac.uk 

- University of Edinburgh, John Levine, 

johnl@aiai.ed.ac.uk 

- University of Huddersfield, TCU Knowledge Engineering, 

Lee McCluskey, 

t.l.mccluskey@zeus.hud.ac.uk 

- University of Manchester, Nikolay Mehandjiev, 

Nikolay.Mehandjiev@co.umist.ac.uk 

- The Open University Walton Hall, Massimiliano 

Garagnani, 

M.Garagnani@open.ac.uk 

- Salford University, TCU Intelligent Manufacturing, 

Ruth Aylett, 

R.S.Aylett@iti.salford.ac.uk 

- Troy Associates Ltd., Vince Long, 

vlong@troyassoc.com 

Associated Members 

- Norman Sadeh, sadeh@cs.cmu.edu, Carnegie 

Mellon University 

- Peter Jarvis, Jarvis@ai.sri.com, SRI 

- Brian Drabble, drabble@cirl.uoregon.edu, 

University of Oregon 

- Sylvie Thiebaux, Sylvie.Thiebaux@anu. 

edu.au, The Australian National University 

The network is open to new nodes at any time.

Newsletter of the European Network of Excellence ... - Service - DFKI

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?