Multi-Agent Cooperative Pursuit Algorithm for UGVs Based on MASAC
Min Fang, Jun Wang · 2025
This study investigates the multi-agent cooperative pursuit problem in complex environments. We overcome traditional algorithm constraints in state information exchange and exploration through decoupled reward function design and state decomposition techniques. The decoupled reward function enables each agent to independently optimize its behavior, while state decomposition reduces state space complexity and improves training efficiency. In addition, our comparative experiments with two neural network models validate the method's effectiveness in enhancing information exchange and exploration performance. Results demonstrate significant advantages of our approach in multi-agent cooperative optimization tasks.