Multiple agents cooperative control based on QMIX algorithm in SC2LE environment
Xingchen Fang, Peng Cui, Qingling Wang · 2020 7th International Conference on Information, Cybernetics, and Computational Social Systems (ICCSS) · 2020
This paper studies the problem of multiple agents cooperative control in some complex scenarios. We use the currently value based multi-agent reinforcement learning algorithm QMIX to realize collaborative control. In this paper, we focus on the solution of traditional centralized training with decentralized execution problem (CTDE) in the multi-agent reinforcement learning tasks and the performance of algorithm in some complex environments. We conduct our experiments in the StarCraft II micromanagement scenarios based on the SC2LE environment, the results verify that QMIX algorithm adopted in this paper can obtain more satisfactory training results compared with other reinforcement learning algorithms and effectively improve the effect of multi-agents cooperative control.