A Method for Reactive Navigation of Nonholonomic Robots in the Presence of Obstacles

Abstract

AbstractWe present a method for guidance of a Dubins-like vehicle towards a target in a cluttered maze-like environment. The vehicle is strongly information and memory limited. In particular, it has no knowledge about the environment and is not capable of memorizing its characteristics. The sensor system provides only the distance to the nearest obstacle if this distance is within the given sensor range, and also gives a partial access to the target relative bearing angle. We examine the simple memoryless static local controller that implements the simple pursuit guidance at a large distance from the obstacles and combines this guidance with collision avoidance activity in a vicinity of obstacles. This activity is undertaken when and only when the distance to the obstacle is decreasing and consists in a maximally sharp turn. At the start of this turn, its direction is randomly chosen; evidence in favor of the random choice option is presented. Mathematically rigorous analysis of this law is provided and it is proved that the vehicle necessarily reaches the destination. Convergence and performance of the proposed controller are confirmed by computer simulations.