An embedded parallel computing architecture for multi-limbed robots

Modern day robots combined with computational intelligence are able to assist humans in increasingly complex tasks. These advanced applications are computationally power hungry to the extent that alternatives to brute-force single core processing are required. Parallelism is a computational technique that allows tasks to be distributed over an array of processors, increasing overall computational ability. This paper presents a generalised framework that allows multi-limbed robots to be deployed using a parallel architecture and secondary data paths. An instance of the parallel architecture is deployed on a hexapod as proof of concept. Here the kinematic equations are represented using Denavit-Hartenberg parameters and solved using the iterative Jacobian transpose. The inverse kinematic equations were distributed across a parallel embedded cluster. The architecture provided a speedup of 5.41 in this application. The proposed architecture facilitates robotic modularity, architectural robustness and critical situational stability.