The authors propose a technique for autonomous robot navigation in terms of a system of concurrent processes running on the computing system of the robot. This system is shown to be free from deadlocks and starvation. Performance analysis of the concurrent processes reveals that the various path planning and learning operations have to be expedited in order to reduce the time of navigation for any path. This necessitates an obstacle terrain model that efficiently supports all the operations involved in path planning and learning. A modified adjacency-list data structure is proposed for the obstacle terrain model and proved to be efficient in supporting the operations to be performed on the model. The complexities of various algorithms for manipulating the terrain model are estimated. These algorithms are currently being implemented on a hypercube computing machine mounted on HERMIES-II at Oak Ridge National Laboratory.
