Intermittent Pursuit of Noncooperative Space Targets Under Discontinuous Observations via Deep Reinforcement Learning

This paper proposes a Deep Reinforcement Learning (DRL)-based framework to develop optimal intermittent pursuit strategies against stochastic, noncooperative space targets under discontinuous observation conditions. Traditional pursuit-evasion models, such as those grounded in differential games, typically rely on continuous observation and control—assumptions that are often invalid in practical space missions due to limited visibility windows and intermittent tracking. These limitations reduce the applicability of classical approaches in real-world scenarios. To address these challenges, we formulate a pursuit-evasion model that explicitly incorporates both discontinuous observations and intermittent control across four representative scenarios. Building on the Deep Deterministic Policy Gradient (DDPG) algorithm, we propose an enhanced method—Intermittent Observation and Control DDPG (IOC-DDPG)—designed to learn robust pursuit policies under realistic constraints. We validate the proposed approach in a noncooperative spacecraft rendezvous case study involving stochastic target maneuvers. Simulation results show that IOC-DDPG consistently achieves a success rate exceeding 70%, outperforming traditional methods across diverse and challenging mission conditions. These results highlight the practical potential of deep reinforcement learning in addressing real-world space mission constraints, and pave the way for future extensions to multi-agent pursuit-evasion scenarios and partially observable environments.