Robust DDPG Reinforcement Learning Differential Game Guidance in Low-Thrust, Multi-Body Dynamical Environments

Onboard autonomy is essential for deep-space missions, where spacecraft must navigate complex, multi-body environments with limited computational resources. Traditional guidance methods often require simplified dynamic models or high computational power, making them impractical for onboard implementation. This study proposes a reinforcement learning-based differential game framework for designing an adaptive, closed-loop controller for low-thrust spacecraft guidance. Unlike conventional methods, this approach does not require an explicit analytical model of the system's dynamics. Instead, it learns directly from the nonlinear motion equations, enabling a data-driven control strategy. A neural network-based controller is trained using the Deep Deterministic Policy Gradient Differential Game (DDPG-DG) algorithm to generate real-time low-thrust control commands efficiently. The method is validated through simulations of Lyapunov orbit transfers in the Earth-Moon system, where it demonstrates strong robustness to disturbances, engine imperfections, and environmental variations. The results highlight the potential of reinforcement learning-based differential game strategies to enhance spacecraft autonomy in complex gravitational environments.

کلیدواژه ها:

Reinforcement learning ، Differential games ، Robust control ، Low-thrust propulsion ، Multi-body dynamics