Presentation PDF - Knowledge Based Systems Group - TU Delft
Presentation PDF - Knowledge Based Systems Group - TU Delft
Presentation PDF - Knowledge Based Systems Group - TU Delft
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Recognising situations in a<br />
flight simulator environment<br />
GAME-ON 2002 Conference<br />
Patrick Ehlert, Quint Mouthaan and Leon Rothkrantz<br />
November 30, 2002<br />
The ICE project<br />
<strong>TU</strong> <strong>Delft</strong><br />
<strong>Delft</strong> University of Technology
Overview of presentation<br />
• The ICE project<br />
• FlightGear simulator<br />
• Explorative data analysis<br />
• <strong>Knowledge</strong> based approach<br />
• Conclusions and results<br />
• Future work<br />
2<br />
<strong>TU</strong> <strong>Delft</strong>
The ICE project<br />
• Intelligent Cockpit Environment (ICE)<br />
• Problem:<br />
Increased automation can result in reduced pilot<br />
situation awareness and information overload<br />
• Solution:<br />
Pilot’s assistant, intelligent interface<br />
3<br />
<strong>TU</strong> <strong>Delft</strong>
The ICE project<br />
• Ultimate goal:<br />
Create system to experiment with intelligent<br />
pilot-vehicle interface<br />
• Subgoals:<br />
• Situation recogniser (SR)<br />
• Pilot workload assessor<br />
• Interface decision-maker<br />
4<br />
<strong>TU</strong> <strong>Delft</strong>
The ICE project<br />
• Purpose of SR:<br />
determine current situation using sensor data<br />
from:<br />
• aircraft<br />
(altitude, height, airspeed etc.)<br />
• pilot<br />
(moving stick, throttle etc.)<br />
• flight plan (expected situations and actions)<br />
• SR can be used as first step to A.I. pilot bot<br />
5<br />
<strong>TU</strong> <strong>Delft</strong>
FlightGear simulator<br />
• Reasons for using FlightGear<br />
• Open source<br />
• Multi-platform<br />
• Extendable<br />
• Realistic (in most situations)<br />
• Multiple planes and flight dynamics models<br />
• XML parameter files<br />
• User friendly (mailing-list list support)<br />
6<br />
<strong>TU</strong> <strong>Delft</strong>
FlightGear simulator<br />
7<br />
<strong>TU</strong> <strong>Delft</strong>
Explorative data analysis<br />
• PCA analysis<br />
• Clustering of data into states<br />
• Elman neural network<br />
• Automatically recognise states<br />
• Predict future states<br />
8<br />
<strong>TU</strong> <strong>Delft</strong>
Experiment data<br />
9<br />
<strong>TU</strong> <strong>Delft</strong>
PCA clustering<br />
10<br />
<strong>TU</strong> <strong>Delft</strong>
PCA path tracking<br />
11<br />
<strong>TU</strong> <strong>Delft</strong>
Future state prediction<br />
12<br />
<strong>TU</strong> <strong>Delft</strong>
Elman neural network<br />
13<br />
<strong>TU</strong> <strong>Delft</strong>
Elman neural network<br />
14<br />
<strong>TU</strong> <strong>Delft</strong>
<strong>Knowledge</strong> based SR<br />
• Real-time, on-line interpretation<br />
• Uses state-transitions transitions and production rules<br />
• Easy to adjust<br />
• Interpretation is transparent<br />
• More detailed situation recognition<br />
• Multiple types of airplanes (XML files)<br />
15<br />
<strong>TU</strong> <strong>Delft</strong>
<strong>Knowledge</strong> based SR<br />
• Recogniser output:<br />
• High-level situation(s)<br />
• Expected actions<br />
• Recognised actions<br />
(start, landing, etc.)<br />
(push throttle, set radar,etc.)<br />
16<br />
<strong>TU</strong> <strong>Delft</strong>
High-level situation: example STD<br />
Startup<br />
airspeed > MAX_SPEED_STANDSTILL<br />
airspeed = FULL_THROTTLE<br />
OR<br />
airspeed > MAX_SPEED_TAXIING<br />
airspeed ><br />
MAX_SPEED_STANDSTILL<br />
Take-off<br />
17<br />
<strong>TU</strong> <strong>Delft</strong>
Expected actions: example rule<br />
• Situation: Dogfight<br />
-> Set master arm<br />
Check HUD<br />
Call on radio<br />
Set IFF off<br />
18<br />
<strong>TU</strong> <strong>Delft</strong>
Results and conclusions<br />
• Airplane sensor data can be clustered with<br />
PCA<br />
• Prediction of simple future states with<br />
Elman neural network possible<br />
• Rule-based system gives excellent and<br />
flexible high-level situation recognition<br />
19<br />
<strong>TU</strong> <strong>Delft</strong>
Future work<br />
• Add pilot actions recognition<br />
• Comparison with flight plan<br />
• Add probability values for reasoning about<br />
concurrent situations/actions<br />
20<br />
<strong>TU</strong> <strong>Delft</strong>