Spacecraft Autonomous Collision Avoidance and Maneuver Based on Deep Reinforcement Learning

Autonomous collision avoidance functionalities are critical technologies for maintaining spacecraft operational safety and enhancing mission success probabilities. Traditional methods rely on heuristic-based approaches or simple reinforcement learning algorithms, which lack the ability to handle complex, dynamic environments with temporal dependencies. In this paper, we present a novel deep reinforcement learning framework, which integrates a gated recurrent unit into the actor network to capture long-term dependencies and enhance decision-making capabilities. Simulations of spacecrafts in low earth orbit show that the proposed algorithm outperforms the traditional deep deterministic policy gradient algorithm in maintaining safe distances and optimizing fuel consumption. Our approach provides a more robust and efficient solution for spacecraft navigation in complex, debris-dense scenarios.