Modeling and Analysis of Cooperative Pursuit Actions in Two-Dimensional Multi-Agent Pursuit-Evasion Games

This study explores pursuit-evasion games in a simulated two-dimensional grid world driven by artificial intelligence strategies, focusing on modeling and analyzing individual and cooperative pursuit actions. Agents are restricted to grid-based movement within defined boundaries, with pursuers aiming to minimize terminal time and the evader seeking to maximize it. In this context, mathematical models are developed for individual and cooperative actions. The cooperative actions, in particular, are formed by combining individual actions based on the agents’ positions and movement directions. Using the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm and a reward mechanism independent of direct distance guidance, agents generate intelligent strategies and extensive pursuit-evasion trajectories. These trajectories are analyzed to identify efficient action patterns at decision points. Key cooperative actions, such as "cooperative driving", "cooperative intercepting" and "encircling" have proven highly effective. Overall, these findings provide a systematic framework for understanding cooperative dynamics in pursuit-evasion scenarios and support strategy development for applications such as Unmanned Aerial Vehicles(UAVs), spacecraft, and autonomous systems.