文献类型：期刊文献

中文题名：多智能体强化学习应对低质量奖励的方法综述

英文题名：A review of methods for multi-agent reinforcement learning in the face of low-quality rewards

作者：Shi, Xin-Hao[1,2]; Yang, Tao[1,2]; Xu, Cheng[1]; Zeng, Qing-Han[2]; Liu, Hong-Zhe[1]; Yang, Yu-Lin[2]

第一作者：Shi, Xin-Hao

机构：[1] Beijing Key Laboratory of Information Service Engineering, Beijing Union University, Beijing, 100101, China; [2] Scientific and Technical Center for Innovation, Beijing, 100012, China

第一机构：北京联合大学北京市信息服务工程重点实验室

年份：2026

卷号：43

期号：6

起止页码：1171-1193

外文期刊名：Kongzhi Lilun Yu Yingyong/Control Theory and Applications

收录：EI(收录号：20263021157281)

语种：中文

外文关键词：Computational methods - Intelligent agents - Reinforcement learning

摘要：Multi-agent reinforcement learning has recently become a research hotspot in the field of reinforcement learning. In reinforcement learning, the reward signal serves as immediate feedback from the environment on the agent’s actions, guiding the agent to learn how to perform better in specific tasks. The quality of the reward signal is crucial for guiding the agent toward learning optimal policies. However, issues such as reward sparsity and credit assignment significantly degrade the quality of the reward signal, posing severe challenges to the learning efficiency and stability of agents. In multi-agent environments, low-quality reward signals further exacerbate these challenges. This paper systematically investigates the causes and implications of low-quality reward signals in MARL and reviews the state-of-the-art methods for addressing these challenges. By providing a comprehensive overview of low-quality reward signals, this paper offers a reference framework for researchers on designing effective learning strategies in environments with low-quality rewards and outlines potential directions for future research. ? 2026, South China University of Technology. All rights reserved.