Deep reinforcement learning based multi-agent non-cooperative game strategy approach

Abstract

To address the problem of deep learning of multi-agent evolution based on cooperative equilibrium strategies without considering non-cooperative interactive self-learning games and evolution, this paper introduces non-cooperative game evolution into the interactive self-learning framework based on the MADDPG algorithm and designs a multi-agent interactive self-learning game evolution method. The experimental results show that the training curve tends to stabilize and reach non-cooperative equilibrium after the training of this method. The visualized experimental results are obtained by reproducing the experimental environment on Ubuntu. It is finally demonstrated that the MADDPG method based on the noncooperative equilibrium strategy has significant learning ability in terms of multi-agent.