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.
This research was supported by the Future Combat System Network Technology Research Center program of Defense Acquisition Program Administration and Agency for Defense Development. (UD190033ED).