Machine Learning and Friends Lunch
Approximate Models, Utility and Robotic Motion Planning
Brendan Burns
UMass
Abstract
The task of motion planning for a robot with many degrees of freedom has proven computational complexity. In order to work around this difficulty, approximate methods are necessary. These approximate methods build approximate models of the robot's configuration space and use these models for motion planning.
This talk will explore the role of approximate models in motion planning and how the use of more sophisticated modeling techniques can aid in the development of more efficient motion planners. We will also discuss the notion of utility and how it can be used (in conjunction with an approximate model) to guide the actions of of a motion planning algorithm so that it maximizes its computational efficiency. Results from several recent experiments and future direction of the work will be presented.