Seminar: 3/16 - David Zipser
11:00 to 12:30 PM at:
Department of Cognitive Science
UC San Diego
"Computing Movement Geometry -- A step in sensory-motor transformations"
Making sensory specified goal directed movements requires the solution of many difficult computational problems. The solution to these problems can be simplified by decomposing the process into a series of stages. In this talk I will show what can be accomplished by a stage that computes the geometrical aspects of movement without regard for time or forces. The output of the geometrical stage provides information that simplifies the task of subsequent stages that deal with the physics of movement. In particular I will describe a computational model of the geometrical stage that uses a relatively simple gradient technique to solve such problems as transformations from extrinsic to intrinsic reference frames, specifying movement paths, removing under-specification due to excess degrees of freedom, and does a considerable amount of constraint satisfaction and error correction. The model is used to simulate the 3D movements of an arm with seven degrees of freedom. Comparing these simulated movements to human movement data reveals the validity of several of the models behavioral predictions.