Learning of plan execution policies for indoor navigation

摘要

Most state-of-the-art navigation systems for autonomous service robots decompose navigation into global navigation planning and local (reactive) navigation. While the methods for navigation planning and local navigation themselves are well understood, the plan execution problem, the problem of how to generate and parameterize local navigation actions from a given navigation plan, is largely unsolved.This article describes how a robot can autonomously learn to execute navigation plans. We investigate how the robot can acquire causal models of the actions executable by the local navigation system and we develop a decision theoretic action selection function which uses the models learned to execute a given navigation plan. Finally, we show, both in simulation and on a RWI B21 mobile robot, that the learned action selection function improves the robot's navigation performance compared to standard plan execution techniques.