Air Collision Avoidance Method of Civil Aircraft Based on Reinforcement Learning

This paper studies on the high-density airspace route planning problem. A Q-Learning algorithm is proposed by considering the time series of states. Furthermore, a N-round loop algorithm is introduced to improve Q-Learning effectiveness for the large action space. Typical scenario simulations are carried out to illustrate the proposed algorithm performance, including a route collision with multiple aircrafts, no-fly zones and static obstacles. The simulation results validate feasibility and effectiveness of the proposed algorithm.