Neural fields for local path planning

Abstract

In this article we introduce a neural field approach for local path planning of an autonomous mobile robot. The robot's heading direction is determined by the localized peak and its velocity by the maximum activation in the field. We emphasize the neural field's ability to keep the path planning stable even in the case of noisy sensor data or varying environments. The theoretical frameworks is validated by an implementation on our mobile service robot called 'ARNOLD'. Since its only sensor is an active stereo camera head, we highlight the importance of gaze control and low-level short-term memory for local path planning, particularly in cluttered indoor environments.