An Adaptive policy Entropy based Action Exploration for Multi-Agents Deep Reinforcement Learning in Mixed-Game

Despite the rapid development of deep learning and reinforcement learning techniques, which have significantly enhanced decision-making capabilities in dynamic environments, the increasing complexity and variability of these environments place higher demands on agents’ exploration skills and adaptability. In modern multi-agents game, interactions extend beyond mere competition or cooperation and instead involve dynamical mixed-game. To address this issue, this paper investigates how agents can determine optimal exploration strategies within such mixed-game contexts and proposes an adaptive policy entropy based multi-agents action exploration method (MEAC), aiming to improving the efficency on reward exploration for multi-agents deep learning in mixed-game. In particularly, in our method, the rewards for game equilibrium, the system utility function and the adaptive policy entropy are involved to collectively enhance learning efficency and coordination for deep reinforcement learning in multi-agents mixed game. Via the exaluations on mixed-game treasure-hunt environments, the experimental results demonstrated that the proposed method outperforms existing approaches in terms of reward acquisition and exploration efficiency.