Integrate multi-agent simulation environment and multi-agent reinforcement learning (MARL) for real-world scenario
Sangho Yeo, Seungjun Lee, Boreum Choi, Sangyoon Oh · 2020
Multi-agent deep reinforcement learning has made a great achievement in deep reinforcement learning through modeling a real-world scenario with multiple agents that communicate with a single environment. However, the test and validation of MARL model on the conventional multi-agent simulation are limited. In this study, we analyze an effective method to use a multi-agent simulation to test and validate multi-agent reinforcement learning models and methods as well as propose two requirements, an intuitive interface and the optimization of simulation, to achieve it.