A GPU Accelerated Path Planner for Multiple Unmanned Aerial Vehicles

Abstract

Unmanned aerial vehicles (UAV’s) have experienced an increased usage in the execution of surveillance and reconnaissance tasks, primary reasons being their versatility, low cost, elimination of human risk, and potential autonomous capabilities. This task requires the aircraft to overfly specified points of interest in an efficient manner whilst avoiding terrain and dangerous regions. To accomplish this autonomously, a path planning module capable of calculating and determining the most appropriate route must be implemented. It must be capable of providing a solution in a robust and timely manner to allow for live flight path updating. This paper proposes a flight planner for a reconnaissance scenario in which multiple UAV’s are required to overfly numerous points of interest (POI) in a given geographical area. The approach in this paper is presented as a three step solution; the set up and formatting of input data, solving the single source shortest point problem for each POI using Bellman Ford, and the distribution and assignment of the appropriate path for each UAV using the Genetic Algorithm. It was shown that the acceleration of this process, achieved by using a Graphics Processing Unit (GPU) allowed for an average speed-up of 11x allowing for rapid path planning.