|
|
Foundations of Robotics
Seminar, November 6, 2007
Time
and Place | Seminar Abstract
Optimal
sampling in the space of paths
Smith hall 100
Talk 4:00 pm
We have been looking for principled techniques for discretizing controls in motion planning by asking whether
and why one strategy for discretization is better
than another. While one can always test
one set of controls against another in a large set of random environments, such
results may not illuminate the fundamental reasons why one may outperform
another. We define the relative completeness of a set of controls
paths) as the probability (over
all worlds) that at least one control will not intersect an obstacle. Likewise,
the relative optimality of a set is the degree to which an optimum solution in
the set exceeds the optimum in the continuum.
We have been investigating two intuitive notions for
their impact on the above metrics. First the dispersion of a set of paths is
related fairly directly to relative completeness. Second, after the choice of a
control, the expected remaining distance to the goal (over all worlds) is
related to the shortest possible path in an obstacle free world.
The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.