Fault tolerant locomotion for walking robots

Abstract

We introduce a general method of planning fault tolerant motion for a robotic task based on the least constraint (LC) framework, which uses a set of constraints on the robot's configuration over time. A performance measure called longevity is defined which, for a given configuration and type of fault, describes the potential for future progress towards the goal. This measure examines the connectivity of the configuration space given failure. An algorithm for computing longevity based on dynamic programming is described. Using the longevity computed at discrete points, a path is computed which is optimally fault tolerant. The set of paths which maximize longevity form a contingency plan for faults occurring at each point. Using LC we specify a gait for a four-legged walking robot. A prototypical step is produced using the longevity measure, which we compare it to a straight-line motion implementation. The optimal longevity paths are shown to be significantly more fault tolerant than the straight-line motion.