|
|
Foundations of Robotics Seminar, February 23, 2010
Time
and Place | Seminar Abstract
Computing Homotopy-Like Equivalence Classes in Hierarchical Motion Planning (a half-baked talk)
Ross Knepper
Robotics Institute
CMU
NSH 1507
Talk 1:30 pm
Homotopy is a topological property on paths that allows us to assign equivalence classes according to what "route" paths take amongst obstacles. In 2D, a robot may choose to drive to the left or right of a traffic cone. Similarly, in 3D a robot may go around a donut or fly through the hole. While there are, topologically, an infinite number of homotopy classes in each example, each case presents two routes that are most useful in making progress toward a goal.
Given a set of precomputed trajectories, I will present an approach to generate these useful equivalence classes of paths in real time. Knowledge of these equivalence classes suggests many promising applications, including collision-testing of some further paths almost for free, increasing robustness to sensor noise in selecting a plan, and increasing safety during execution. After describing these benefits, I will do a (half-baked) explanation of the theoretical fundamentals and solicit comments from the audience on how to generalize the technique for robots operating in diverse configuration- and work-spaces.
The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.