An Optimised GPU-Based Robot Motion Planner

This paper presents a novel approach for solving Dynamic Programming optimisation problems such as the generation of exhaustive costto-go functions for path planning in dense environments. Cost-to-go function generation on an occupancy grid is the core of many metricbased techniques for unmanned ground and aerial vehicle path planning. The main limitations of existing methods concern a trade o between properties such as physical size of occupancy grid, resolution of grid cells and required update frequency. This paper presents a concurrent version of a traditional dynamic programming algorithm used to evaluate a cost function on grid-based domains. The proposed algorithm provides an existing system with an order of magnitude more exibility and leniency in the aforementioned properties as it provides mathematically equivalent results to the traditional algorithm in an order of magnitude less time on currently available hardware. Although this algorithm has been developed and tested in a robotics context, it may benet other areas of optimisation and control.