Yifan Zhong, Jakub Grudzien Kuba, Siyi Hu, Jiaming Ji, Yaodong Yang
TL;DR本文提出了基于 HARL 算法的新框架 HAML,将多智能体强化学习的合作扩展到异构智能体模式,并对该框架下的多种算法进行了验证和比较。测试表明,HARL 算法在协调异构智能体方面的稳定性和有效性要优于现有的 MA 对应物。
Abstract
The necessity for cooperation among intelligent machines has popularised
cooperative multi-agent reinforcement learning (MARL) in AI research. However,
many research endeavours heavily rely on parameter sharing a