GPU based approach for fast generation of robot capability representations

Capability maps are an important tool for enabling robots to understand their bodies by providing a way of representing the dexterity of their arms. They are usually treated as static data structures be- cause of how computationally intensive they are to generate. We present a method for generating capability maps taking advantage of the parallelization that modern GPUs offer such that these maps are generated approximately 50 times faster than previous implementations. This system could be used in situations were the robot has to generate this maps fast, for example when using unknown tools.