Collision Avoidance for Unmanned Aerial Vehicles using Simultaneous Game Theory

It is predictable that future applications for unmanned aerial vehicles will have a high degree of automation and will have to be integrated in civil airspace in order to go beyond line-of-sight. Nevertheless, unmanned aircraft are still unable to detect obstacles and to avoid collisions, resulting in a significant operational risk. This paper aims to introduce a new game theoretical approach for collision avoidance in order to address this gap. The algorithm builds a simultaneous game between two players, which is then solved using game trees. The own vehicle and the intruders are the players in the game and have different possible moves. A cost function is used to describe the consequences of the players'actions for each combination of actions they perform. The introduced collision avoidance algorithm was implemented and integrated into the X-Plane flight simulator with a large number of generated simulations scenarios, covering test cases like multiple intruders on a direct collision course or densely populated airspace with irregular traffic. The proposed algorithm is real-time capable, which is important for further use as part of a collision avoidance system.