Hunan University | robust Multi-Agent Reinforcement Learning in noisy environment

【 title 】Robust multi-agent reinforcement learning for noisy environments

【 The author team 】Chen, Xinning, Liu, Xuan, Luo, Canhui, Yin, Jiangjin

【 Date of publication 】2022.1.22

【 Thesis link 】https://link.springer.com/content/pdf/10.1007/s12083-021-01133-2.pdf

【 Recommended reasons 】 Despite recent intensive learning (RL) Progress has been made in , But by RL Trained agents are usually sensitive to the environment , Especially in multi-agent scenarios . The existing Multi-Agent Reinforcement learning methods can work well only under the assumption of perfect environment . However , The real world environment is usually noisy . Inaccurate information obtained from noisy environment will hinder the learning of agent , Even lead to training failure . This paper focuses on the problem of training multiple robust agents in noisy environment . In this paper, a new algorithm is proposed , Multi-agent fault-tolerant reinforcement learning （MAFTRL）. The main idea of this paper is to establish the error detection mechanism of agent itself , Design the information communication medium between agents . The error detection mechanism is based on automatic encoder , Calculate the reliability of each agent's observation , Effectively reduce environmental noise . Communication media based on attention mechanism can significantly improve the ability of agents to extract effective information . Experimental results show that , The method in this paper accurately detects the error observation of agents , It has good performance and strong robustness in traditional reliable environment and noisy environment . Besides ,MAFTRL It is obviously superior to traditional methods in noisy environment .