Dynamic issues for mobile robot real-time discovery and path planning

Abstract

Sensor-based discovery path planning is the online guidance of a discovery agent (robot) who begins execution without a complete a priori map. The presented approach is locally optimal in the sense of minimizing the hitting probability while minimizing the distance to the goal based on what is currently known about the world. The technique is implemented as a process-based client-server. The integration and relationship between different facets of the algorithm are explored in terms of their effect on the real-time performance of the robot. The computation time for the minimal hitting probability surface is reduced by using methods of relaxation and a non-uniform sampling grid. The path is executed by a separate process reading gradient values off of the computed surface at regular intervals. Sensor updates are input as a collection of geometric primitives using a simple interface. Current work is focused on incorporating depth information from passive visual sensors.