arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.13708cs.LGcs.AIcs.MA
收藏

MARLlib:一个可扩展且高效的多智能体强化学习库

MARLlib: A Scalable and Efficient Multi-agent Reinforcement Learning Library

ReLER, AAII, University of Technology Sydney(悉尼科技大学);Institute for Artificial Intelligence, Peking University(北京大学人工智能研究院);NetEase Fuxi AI Lab(网易伏羲人工智能实验室);MBZUAI(穆罕默德·本·扎耶德人工智能大学);School of Intelligent Systems Engineering, Sun Yat-sen University(中山大学智能工程学院);Shandong Artificial Intelligence Institute, Qilu University of Technology(齐鲁工业大学山东省人工智能研究院)

Siyi Hu, Yifan Zhong, Minquan Gao, Weixun Wang, Hao Dong, Xiaodan Liang, Zhihui Li, Xiaojun Chang, Yaodong Yang

AI总结:

针对多智能体强化学习领域缺乏兼容高效开发库的问题,研究人员提出MARLlib库,通过标准化环境封装、智能体级算法实现和灵活策略映射三大机制解耦任务与算法,实现自动适配训练策略,且代码已开源。

摘要:

多智能体强化学习(MARL)领域研究人员面临的一个重大挑战,是找到一个既能为多智能体任务与算法组合提供快速兼容的开发环境,又无需考虑兼容性问题的库。本文提出MARLlib库,通过三大核心机制解决上述挑战:1)标准化的多智能体环境封装器,2)智能体层面的算法实现,3)灵活的策略映射策略。借助这些机制,MARLlib能有效解耦多智能体任务与算法学习过程的交织特性,还可根据当前任务属性自动调整训练策略。MARLlib库的源代码已在GitHub公开:https://github.com/Replicable-MARL/MARLlib。

英文摘要:

A significant challenge facing researchers in the area of multi-agent reinforcement learning (MARL) pertains to the identification of a library that can offer fast and compatible development for multi-agent tasks and algorithm combinations, while obviating the need to consider compatibility issues. In this paper, we present MARLlib, a library designed to address the aforementioned challenge by leveraging three key mechanisms: 1) a standardized multi-agent environment wrapper, 2) an agent-level algorithm implementation, and 3) a flexible policy mapping strategy. By utilizing these mechanisms, MARLlib can effectively disentangle the intertwined nature of the multi-agent task and the learning process of the algorithm, with the ability to automatically alter the training strategy based on the current task's attributes. The MARLlib library's source code is publicly accessible on GitHub: \url{https://github.com/Replicable-MARL/MARLlib}.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑