Chapter 2. Prehension

Chapter 8 - Constraints on Human Prehension 323
etc.). Affecting the choice of an opposition space, intrinsic and
extrinsic muscles make differential contributions to movement, while
muscle, tendon, and joint receptors provide information about results
of those movements. The existence of low level sensorimotor features
-- such as the tonic vibration reflex, pad alignment in opposition,
coupled degrees of freedom in the fingers, ligaments being used for
passive control, and rapid grip force adjustments -- and higher level
neural control, such as active control of muscles and the pyramidal
tract for fractionated finger movements are noted.
Opposition space as a model takes into account the hands ability
to be both an input and output device, dealing with applying task
forces while gathering sensory information. It addresses the hand in
terms of its oppositional capabilities, providing combinations of
oppositions and VF mappings that match the requirements of the task.
Two important benefits to using a Marr type view for prehension are
observed. On one side, it separates implementation details from a task
description, a trend occurring in programming languages in general,
and robot programming languages in particular. This allows the
functional study of human hand to be carried over to dextrous robot
hands. The mapping from opposition space into the human hand,
with its particular physical/biomechanical and its sensorimotor
constraints, can be replaced with a robot hand that has its own
mechanical and motor constraints (Figure 8.1, right side). The human
Central Nervous System is replaced by a Computational Nervous
System, comprised of an expert system, feedforward and feedback
controllers, and/or software simulating a neural network. Or perhaps
it could be a hardware implementation, such as a distributed network
of computers, transputers, RISC processors, or even neural network
hardware. The mapping from actual to robot hand changes without
redoing the overall high level description of hand functionality. Of
course, until a robot has a reason to grasp an object, other than being
told to do so, the only high level effects would be functional ones.
The advantage of this model is that movement occurs satifying all
the constraints acting on the system. Goals from the upper level filter
down while the constraints from the biomechanics and anatomy filter
up. This allows an opposition space description to be device
independent, and therefor, the postures of the human hand can map
onto other manipulators. Motor commands are generated at three levels
(opposition space, biomechanical, sensorimotor) and act within a
constraint space of possibilities.
Another issue is the goal to understand the versatility of human
prehension in general. With a large repertoire of available movements,