Login
Computer Science @ Rochester
Wednesday, April 28, 2004
11:00 AM
CSB 209
Ph.D. Thesis Proposal
Xue Gu
University of Rochester
Movement Control Model for Articulated Agent
Robotics has been successfully applied to industry for several decades. However, to develop really adaptive, dexterous and autonomous real-life robots, there is still a long way to go. One difficulty is that currently we still do not fully understand how the human brain encodes and decodes motor control. There are two groups of approaches dedicated to motor control. The top-down approaches are simple and static, but is not adaptive and scalable to systems with high Degrees of Freedom (DOF). An alternative is the bottom-up methods, inspired from biological results. Motion routines would encode basic actions, and higher-level complex behaviors can be generated by superposition or combination of a repository of routines. While the bottom-up approaches are dynamic and adaptive to high DOF system, how the motion routines are represented and coupled for composite behaviors are still under research. Our ultimate goal of this proposal is to build up a computational model that can control a high DOF anthropomorphic agent to produce realistic movements in a natural way. With this goal, we propose to use top-down approaches within the motion routines, and the bottom-up methods to generate of more complex whole body movements by activating multiple routines with different parameters. In modeling one general motion routine of reaching, we attempt a kinematics computational model with dynamics optimization to determine the end posture before the movements actually start. In the execution phase, a spring-muscle model based on the well-known equilibrium theory is applied to the arm to actually control the movements.