A computational feasibility study of failure-tolerant path planning

This work considers the computational costs associated with the implementation of a failure-tolerant path planning algorithm proposed in [1]. The algorithm makes the following assumptions: a manipulator is redundant relative to its task, only a single joint failure occurs at any given time, the manipulator is capable of detecting a joint failure and immediately locks the failed joint, and the environment is static and known. The algorithm is evaluated on a three degree-of-freedom planar manipulator for a total of eleven thousand different scenarios, randomly varying the robot's start and goal positions and the number and locations of obstacles in the environment. Statistical data are presented related to the computation time required by the different steps of the algorithm as a function of the complexity of the environment.