Coordination Between Individual Agents in Multi-Agent Reinforcement Learning

The existing multi-agent reinforcement learning methods (MARL) for determining the coordination between agents focus on either global-level or neighborhood-level agents. However problem of individual is remain to be solved. It crucial an optimal coordinated policy in unknown environments analyze agent's roles and correlation To this end, paper we propose agent-level based MARL method. Specifically, it includes two parts our first analysis Pearson, Spearman, Kendall coefficients; And second training framework where communication message weakly correlated dropped out, a reward function built. proposed method verified four mixed cooperative-competitive environments. experimental results show that outperforms state-of-the-art can measure accurately.