A symbiotic simulation architecture for evaluating UAVs collision avoidance techniques

Unmanned aerial vehicles (UAVs) currently have many real-world applications such as surveying, delivering small packages, and military applications. In order to maintain safe flight, UAVs require collision avoidance to divert them from collisions or other known dangerous paths. Many algorithms have been developed and implemented, but seldom are they compared to each other analytically within a single system. In this paper, an implementation of a symbiotic simulation architecture is described to conduct cyber-physical experiments and to allow different collision avoidance algorithms to be tested in real-time. Three algorithms based on mixed-integer linear programming (MILP), sparse A* search, and artificial potential fields (APF) are implemented, and results of computational experiments are presented. MILP provides the most efficient paths when given enough time. A* and APF exhibit similar performance, with APF being the least computing expensive. On the basis of the results of the experiments, three hybrid algorithms are proposed for future tests using the embedded simulation architecture.