Abstract
This paper designs optimal control polices for networked multiagent pursuit-evasion game (MPEG) problems based on reinforcement learning (RL) technique. Depending on the number of evaders, MPEG is formulated into several simpler multiple-pursuer single-evader games (MPSEGs) by a divide and conquer approach. Then we propose optimal control policies for all the agents in each MPSEG, which constitute a distributed Nash equilibrium, and provide the capturability and Nash equilibrium analysis. Finally, a data-driven RL algorithm is developed to online learn optimal control polices using measurable behavior data. A simulation example is given to verify the effectiveness of the proposed approach.
Funding Information
  • National Key R&D Program of China (2018YFA0702200)
  • National Natural Science Foundation of China (62203208, 62373196, 52077106)
  • Natural Science Research Project of Jiangsu Higher Education Institutions (20KJB120002)

This publication has 34 references indexed in Scilit: