The navigation system of the JPL robot

Abstract

The control structure of the JPL research robot and the operations of the navigation subsystem are discussed. The robot functions as a network of interacting concurrent processes distributed among several computers and coordinated by a central executive. The results of scene analysis are used to create a segmented terrain model in which surface regions are classified by traversibility. The model is used by a path planning algorithm, PATH, which uses tree search methods to find the optimal path to a goal. In PATH, the search space is defined dynamically as a consequence of node testing. Maze-solving and the use of an associative data base for context dependent node generation are also discussed. Execution of a planned path is accomplished by a feedback guidance process with automatic error recovery.