Navigation With Uncertainty: II. Avoiding High Collision Risk Regions

Abstract

We present a probabilistic method for sensor based robotic navigation in dynamic and noisy environments. The method generates a trajectory that guarantees a tolerable associated collision risk. Estimates of the obstacle's kinematic parameters and measures of confidence in these estimates are used to produce regions where the probability of encountering any obstacle is bounded by a predefined value. The support of the Defense Advanced Research Projects Agency (ARPA Order No. 6350) and the U.S. Army Engineer Topographic Laboratories under Contract DACA76-89-C-0008 is gratefully acknowledged, as is the help of Sandy German in preparing this paper. 1 Introduction 1.1 Methods in Probabilistic Navigation We are studying probabilistic approaches to sensor-based navigation in noisy dynamic environments. Here, we address the general problem of trajectory planning where an autonomous robot must reach a moving goal point and avoid surrounding moving obstacles. The motions of the obstacles are ...