Anti-collision Trajectory Planning for Satellite Formation Reconstruction Based on Deep Reinforcement Learning

This paper proposes an optimal trajectory planning method for satellite formation reconstruction based on deep reinforcement learning. To begin, the action space, state space, and reward function of satellite formation reconstruction are created, with the collision avoidance constraint taken into account. Second, the algorithm's essential parameters' learning rate and appropriate noise are determined. What's more, the Unity program is developed to create the training environment, so the real satellite dynamics model are embed in the environment. The optimal trajectory of formation satellite reconstruction obtained by this method can better meet the constraints such as collision avoidance, and the calculation speed is fast, which makes the autonomous real-time reconstruction of formation satellite possible. Finally, 1 simulation example is carried out to verify the proposed algorithm, showing that the formation reconfiguration task can be executed successfully while achieving rapid convergence.