Modeling of real-world biological multi-agents is a fundamental problem in various scientific and engineering fields. reinforcement learning (RL) is a powerful framework to generate flexible and diverse behaviors in cyberspace; however, when modeling real-world biological multi-agents,