|
|
Foundations of Robotics
Seminar, May 13, 2008
Time
and Place | Seminar Abstract
ICRA Practice Talks:
- Path and Trajectory Diversity: Theory and Algorithms (Ross A. Knepper)
- Iterated Filters for Bearing-Only SLAM (Stephen Tully)
Ross A. Knepper and Stephen Tully
Newell Simon Hall 1507
Talk 4:30 pm
Path and Trajectory Diversity: Theory and Algorithms (Ross A. Knepper)
A common approach to motion planning for nonholonomic
robots is to construct, offline, a representative set of
paths which are feasible for the vehicle to follow. During
planning, each of these paths is tested against the local map
for collision with obstacles, with the winning path plan
being selected from among the survivors. A limited number of
paths can be evaluated due to a restricted planning time
budget. Not all path sets of a given size are equally
effective in planning. This talk examines one metric, which
we call diversity, for improving planning success rates. A
path set's diversity is proportional to the probability that
at least one path survives. The diversity measure can be
used to design a path set by sampling a path subset of
greatest diversity from a larger set. An exact diversity
computation is exponential in cost, making it impractical for
real problems. Instead, two polynomial-time greedy
algorithms for path set construction will be presented and
analyzed. In comparison to randomly sampled path sets, up to
2.8 times as many diverse paths survive in experimental test
environments.
Iterated Filters for Bearing-Only SLAM (Stephen Tully)
This paper discusses the importance of iteration when performing the measurement update step for the problem of bearing-only SLAM. Traditionally, the extended Kalman filter (EKF) has been used for SLAM, but the EKF measurement update rule can often lead to a divergent state estimate due to its inconsistency in linearization. Instead, we represent the bearing-only update as a numerical optimization problem (solved with an iterative approach such as Gauss-Newton minimization). This prevents divergence of the Kalman filter state and produces accurate SLAM results for a bearing-only sensor. Two outdoor mobile robot experiments are discussed to compare algorithm performance.
The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.