<p>This paper proposes an online learning-based dynamic event-triggered pursuing strategy for multi-satellite orbital pursuit-evasion games with relative state and input constraints. Initially, barrier functions are introduced to enforce state constraints, while a specially designed non-quadratic value function addresses control input saturation. Accordingly, the optimal game-theoretic strategies are proposed and proven to satisfy the Nash equilibrium conditions. To alleviate the onboard communication burden, a dynamic event-triggered mechanism is developed to update the control input of each pursuer only when necessary. The Hamilton-Jacobi-Isaacs equation is reformulated via integral reinforcement learning and solved with a critic neural network to approximate the Nash equilibrium value function without full system dynamics. Furthermore, experience replay and singular value decomposition stack techniques are integrated into the critic’s weight update to avoid the persistency of excitation and improve data efficiency. The uniformly ultimately bounded stability of the system and the weight estimation error are rigorously established via Lyapunov theory. Finally, numerical simulations and full-physical ground experiments jointly validate the theoretical results and demonstrate the superior performance of the proposed method.</p>