Distributed Conflict Resolution at High Traffic Densities with Reinforcement Learning

Future operations involving drones are expected to result in traffic densities that orders of magnitude higher than any observed manned aviation. Current geometric conflict resolution (CR) methods have proven be very efficient at relatively moderate densities. However, densities, performance is hindered by the unpredictable emergent behaviour from neighbouring aircraft. Reinforcement learning (RL) techniques often capable identifying emerging patterns through training environment. Although some work has started introducing RL resolve conflicts and ensure separation between aircraft, it not clear how employ these with a number whether can compare or even surpass current CR methods. In this work, we an method for distributed resolution; completely responsible guaranteeing minimum all aircraft during operation. Two different action formulations tested: (1) where controls heading, speed variation; (2) speed, altitude variation. The final safety values directly compared state-of-the-art algorithm, Modified Voltage Potential (MVP) method. Although, overall, as MVP reducing total losses separation, its actions help identify favourable avoid conflicts. more preventive behaviour, defending advance against nearby yet conflict, head-on while intruders still far away.