Directionality Reinforcement Learning to Operate Multi-Agent System without Communication