Monday Nov 1, WeH 4601, 3:00


		Parameterization as Path Planning for
			  Robot Manipulators

		  	   Goang-Tay Hsu


This is a talk about the preliminary attempt to make path planning for
robot manipulator a more learnable problem.  Traditional path planning
(or trajectory generation) for robot manipulator involves dense sampling
and interpolation.  There is then heavy traffic between the commanding
agent and the execution agents.  This makes it a less viable option as
means for biological motor control, especially for fast motions.  Also,
it is not obvious how to learn and generalize the experiences so that
a similar task can be achieved with less effort and/or more precision.

A parameterization-as-path-planning scheme is introduced as a solution
to the above problems.  The execution agents are assumed to bear
plans for executing parametric functions.  The commanding agent then
need only to provide the parameters along with the indication of which
function to use.  Both redundant and non-redundant planar manipulators
are investigated under this scheme for generalization.