GEO Spacecraft Pursuit–Evasion Under Debris Constraints: A Fused Fuzzy Deep Reinforcement Learning Framework

This article addresses the problem of impulsive orbital pursuit–evasion games in Geostationary Earth Orbit under the constraint of space debris avoidance. To tackle this challenge, a fused fuzzy deep reinforcement learning (FFDRL) framework is proposed. First, a comprehensive modeling framework is established that integrates impulsive maneuver dynamics, debris avoidance, fuel limitations, and mission time constraints, formulating the problem as a constrained Markov decision process. Second, the core FFDRL framework is developed by embedding fuzzy reasoning mechanisms into deep neural networks, forming a fused fuzzy deep neural network architecture within both policy and value networks to improve local interpretability and generalization. The learning process employs the proximal policy optimization algorithm combined with an asynchronous self-play training mechanism, which iteratively optimizes the strategies of both the pursuer and the evader in a two-agent adversarial setting. A multiobjective reward function is carefully designed to balance orbital interception, debris avoidance, and energy consumption, guiding the learned policies toward effective and safe solutions. Finally, simulation studies in planar and 3-D settings demonstrate the effectiveness of the proposed FFDRL framework. Under variations in debris distribution, initial separation, and pursuer maneuverability, the learned policies remain robust and adaptable, indicating that FFDRL provides an effective and practical approach for autonomous spacecraft pursuit and evasion control in complex orbital environments.