Chapter 2. Prehension

0 pposi t ion
vector
Chapter 4 - Planning of Prehension 99
Two networks of 'neurons' as
control surfaces in task centered
coordinate frame
VFl's view VF2's view
Figure 4.14. Iberall Approach Vector Selector model. The
coordinate frame for the computation is task centered around the
opposition vector. On the right are shown the two competing
networks. Local computations occur at each neuron, receiving
excitatory inputs from its four neighbors, from its counterpart in
the other network, and from the inputs to the system. Global
inhibition keeps the computation below a certain level of
excitation.
From measurements taken from the hand of one subject and from
estimates on the range of the degrees of freedom of the hand
articulators (Kapandji, 1982), a kinematic solution set was calculated
and the number of solutions at various wrist locations counted. These
solutions were where normals from the pads of VF1 and VF2 were
equal in value and opposite in direction and also collinear with the
opposition vector. Using the count of potential solutions as input, the
Approach Vector Selector computed a location to put the wrist at time
of contact. In Figure 4.15, a time course trace of the Selector starts
from an initial state and finishes when it has converged on a solution
for the wrist position relative to the object. The network of cells is
initialized to a zero excitation level, as seen in Figure 4.15a. On the
left, the excitatory field VF1 is seen; on the right, the field VF2 is
seen. Simulated activity at 11 time-constants after initialization is seen
in Figure 4.15b, and at convergence in Figure 4.15~ at 27 time-constants
after initialization. The peak formed indicates a reasonable
(within the constraints modelled) and desired solution for the location
of the wrist. Iberall found that higher weights on VF1 were needed
than those on VF2 in order to get convergence. This supports the
behavioral results showing the thumb being more constrained (see