Randomized belief-space replanning in partially-observable continuous spaces

Abstract

We present a sample-based replanning strategy for driving partially-observable, high-dimensional robotic systems to a desired goal. At each time step, it uses forward simulation of randomly-sampled open-loop controls to construct a belief-space search tree rooted at its current belief state. Then, it executes the action at the root that leads to the best node in the tree. As a node quality metric we use Monte Carlo simulation to estimate the likelihood of success under the QMDP belief-space feedback policy, which encourages the robot to take information-gathering actions as needed to reach the goal. The technique is demonstrated on target-finding and localization examples in up to 5D state spacess. © 2010 Springer-Verlag Berlin Heidelberg.