Aircraft collision avoidance modeling and optimization using deep reinforcement learning

초록

We propose an imitation-type reinforcement learning approach for aircraft collision avoidance problems. The policy model is initially supervised to learn the collision avoidance strategies based on the domain-knowledge from the flight mechanics and the guidance contexts, and then it is updated and optimized via reinforcement learning and the proximal policy optimization. The performance of the proposed approach was verified via Monte-Carlo simulation runs that contain a wide range of collision geometries. © ICROS 2021.

키워드