Abstract:
Pattern recognition strategies have been explored for prosthetic hand control for over 20
years, but have not been commercially implemented. [1] Articulated research hands now rival
human hand articulation, presenting new control challenges. Because no methods of pro-
cessing signals of any source (cortical, peripheral nerve or myoelectric) are currently capable
of delivering joint angle intent, the information content of user intent must be expanded.
Control information for the 15 to 20 joints of a highly articulated hand must be derived from
a single variable of grasp information. Two strategies are proposed for augmenting intent to
control these hands. First, an array of grasp endpoints composed of the joint angles desired
is created for each grasp, and the linear path to the desired endpoint constantly updated.
Finger collisions are supervised by the user. Second, a larger set of grasp control parameters
are explored. Using additional parameters allows more precise control of the arrangement of
the ngers than whole hand grasps, a greater degree of intuitive control over the arrangement
of the ngers, and the possibility of dextrous manipulation.
