Chapter 2. Prehension

Chapter 4 - Planning of Prehension 67
our hammer example, before the hammer is placed on the shelf, it
must be picked up; before it is grasped, the hand must first open up.
Successive movements are affected by earlier movements in a se-
quence. As well, earlier movements show anticipatory planning for
subsequent ones. There is no guarantee that each sub-task controller
will be able to complete its sub-task, and thus this plan is open-loop.
The sequence of events is predetermined and applied regardless of
errors. At the same time, the details of what is occurring within each
sub-task controller are not relevant to the overall plan. In fact, the task
planner constructed the plan as a skeleton of calls to sub-task
controllers. Each sub-task controller gets filled in and performs its
piece of the work towards achieving the overall goal. Details about the
frame of reference, the type of movement, and the type of control to
use are expanded upon and then performed. As we saw in the fine
motion example, a sub-task controller is constructed that uses sensory
feedback to perform a guarded motion because contact with the
environment is occurring.
4.2.2 Distributed processing task plans
The Arbib (1985) coordinated control program (CCP, see Figure
3.3) is a task plan using schemas for sub-task controllers. Just as in
the LAMA example, each schema in the plan is involved in some as-
pect of the overall task. However, the LAMA plan was written on a
computer in a modular, but serial format. The CCP is a parallel dis-
tributed processing model for how the CNS might function. For a
given task, all motor schemas in the model are instantiated at the same
time and control lines (dashed lines in Figure 3.3) are activated. When
critical data are received or some processing has occurred, the schema
starts performing its sub-task. Thus, serial order in a CCP is achieved
by the use of activation signals and the passing of control parameters
and data from perceptual to motor schemas2. Order is specified with-
out a central representation. The actual working of each motor
schema, again as in LAMA, is at another level of detail and can be
modelled depending on what is known about control in the CNS.
For a task of hitting a nail with a hammer, not only must the ham-
mer be grasped correctly for the task, it must also be brought to bear
squarely against the head of the nail. Is error handling part of the
2Schema programming has been done in the robot schema language RS (Lyons
and Arbib, 1989). RS formalizes schemas as port autamata that can be instantiated
and deinstantiated.