Using attractor dynamics to control autonomous vehicle motion

Abstract

Planning collisionless paths for autonomous vehicles is a basic task of autonomous robotics. A large body of work has addressed this issue by starting out from symbolic representations of the environment, in which obstacles and targets are represented metrically. In practice, these representations are difficult to obtain, however. This paper shows that the dynamic approach to path planning (Schoner, Dose, 1992) can be implemented on simple autonomous vehicles using only low-level sensory information such as distances sensed by infrared or sonar detectors. Each sensor is assumed to contribute a repulsive force-let to a dynamical system, which generates a trajectory of heading direction. The repulsion force is a decreasing function of sensed distance. This leads to obstacle avoidance. Movement toward a target is achieved by adding an attractive force-let. This architecture generates smooth collisionless trajectories based on real-time sensory information.