A Distributed Simulation-to-Reality Transfer Framework for Multi-agent Reinforcement Learning

Recent works have demonstrated the applications of multi-agent reinforcement learning (MARL) algorithms in many fields, including autonomous driving, finance, and games. However, how to transfer the MARL models trained in the simulation environment to the real world is a challenging issue due to the gap between simulation and reality. To address this issue, we propose a distributed simulation to reality (Sim2Real) transfer framework that encompasses three components, including a platform, MARL algorithms, and a distributed multi-agent Sim2Real transfer algorithm. In addition, to demonstrate the effectiveness of the proposed framework, we design an experimental collaborative hunting task. The experiment results indicate that the proposed framework can enhance the real-world performance of MARL models.