Reinforcement Learning for the Interception of Hypersonic Vehicles

This paper explores using reinforcement learning for the guidance of an interceptor protecting a stationary target against an attacking hypersonic vehicle. The performance of the reinforcement learning-trained algorithm is compared to a traditional guidance method for interception. Proximal policy optimization (PPO) is the algorithm selected to guide the interceptor, and proportional navigation is the chosen traditional method that has been historically implemented. Both methods provide acceleration commands to the interceptor, while the attacker maintains a constant velocity and linear trajectory towards the target. PPO uses an actor-critic framework and samples the environment for an action based on the current policy. Although further work is required for the PPO algorithm to improve upon the performance of traditional methods, it may provide more effective and efficient interception against a highly maneuverable attacking vehicle.