Pursuit–Evasion Game of Multiagent Systems With Limited Perception and Communication

This paper investigates a multi-pursuer single-evader game in multi-agent systems with unknown nonlinear dynamics. Unlike most existing studies that rely on complete information, this work addresses a more realistic scenario where pursuers face incomplete perception and limited communication ranges. To address these challenges, a novel performance index is formulated based on distributed relative distance errors, augmented with a communication cost term that penalizes excessive inter-pursuer separation, thereby promoting reliable information exchange among pursuers. A synchronous integral reinforcement learning-based actor–critic framework is then developed, which learns both the optimal value function and control strategies online directly using trajectory data, eliminating the need for prior dynamics knowledge or stabilizing initial strategies. Rigorous analysis establishes sufficient conditions for saddle-point equilibrium and guarantees uniform ultimate boundedness of the closed-loop errors with an explicit bound. Simulation results based on an unmanned aerial vehicle pursuit–evasion scenario demonstrate the effectiveness of the proposed approach.