MAPPO and Auction Algorithm Based Multi-Target Hunting Algorithm Design and Simulation

This is the authors' abstract. We don't add key points for this paper.

This paper studies the problem of multi-target hunting in a bounded continuous two-dimensional space, where the speed of evaders is supposed to be higher than pursuers. Here, we propose a multi-target hunting algorithm based on the MAPPO (Multi-Agent Proximal Policy Optimization) reinforcement learning algorithms and the auction algorithm (MAPPO-A), and a novel reward function design to guide agents learning. Firstly, we build the kinematics model of pursuers and evaders, and describe the problem of multi-target hunting. Secondly, in order to achieve the efficient target hunting, the auction algorithm is adopted to allocate the evader to the pursuer, where a target allocation model is built. After the target allocation, we further model the pursuit process as the MDP (Markov Decision Process (MDP)), where the state space, action space and reward function with respect to pursuers are defined. Further, the MAPPO algorithm is used to solve the problem. Finally, we give a simulation to validate that this method can complete the multi-target hunting task efficiently.