A swarm-independent behaviors-based orbit maneuvering approach for target-attacker-defender games of satellites

The target-attacker-defender gaming decision problem for satellites with impulse-thrust orbit maneuvering capability only is studied in this paper. A swarm-independent behaviors-based orbit maneuvering approach is proposed. The satellite maneuvering game problem is first transformed into an optimization problem involving impulse size, maneuvering type, and task objectives. A deep reinforcement learning algorithm is employed to optimize this problem. Specifically, eight swarm-independent behaviors are proposed to guide pulse size selection, involving at least 12 parameters related to the initial orbital states of both sides. Additionally, three auxiliary guidance mechanisms are introduced to reduce the optimization space. Finally, fast, autonomous, and stable game maneuvering is achieved. Unlike the distance-based approaches, the proposed method uses process guidance, incorporating more gaming information and constraints. This leads to a more precise training objective and improved training accuracy. Simulation results show that the success rates of the proposed method are over 11% higher than those achieved by distance-based methods in six versus two target-attacker-defender games.