arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2603.02604cs.LG
收藏

异质智能体协作强化学习

Heterogeneous Agent Collaborative Reinforcement Learning

Beihang University(北航);Bytedance China(字节跳动中国);Tsinghua University(清华大学);Peking University(北京大学);Apple(苹果公司)

Zhixia Zhang, Zixuan Huang, Gongxun Li, Huaiyang Wang, Chengyi Yuan, Xin Xia, Deqing Wang, Fuzhen Zhuang, Shuai Ma, Ning Ding, Yaodong Yang, Jianxin Li, Yikun Ban

AI总结:

本文提出了一种新的强化学习从可验证奖励(RLVR)问题HACRL,通过异质智能体共享验证的轨迹实现协同优化,解决了孤立多智能体在线优化的效率问题,并提出HACPO算法以最大化样本利用率和跨智能体知识转移。

摘要:

我们引入了异质智能体协作强化学习(HACRL),一种新的强化学习从可验证奖励(RLVR)问题,旨在解决孤立多智能体在线优化的低效问题。HACRL允许独立执行的协同优化:异质智能体在训练期间共享验证的轨迹以互相改进,而在推理期间独立操作。不同于基于大语言模型的多智能体强化学习(MARL),HACRL不需要协调部署,也不同于在线/离线策略蒸馏,它使异质智能体之间实现双向相互学习,而非单向的教师到学生转移。基于此问题,我们提出HACPO,一种协作RL算法,能够通过原则性的轨迹共享最大化样本利用率和跨智能体知识转移。为缓解能力差异和策略分布偏移,HACPO引入了四个定制机制,具有对无偏优势估计的理论保证。在多样化的异质模型组合和推理基准上的广泛实验表明,HACPO一致地提升了所有参与智能体,相比使用双轨迹的GSPO,平均提高了3.6%,同时仅使用一半的轨迹成本。

英文摘要:

We introduce Heterogeneous Agent Collaborative Reinforcement Learning (HACRL), a new Reinforcement Learning from Verifiable Reward (RLVR) problem that addresses the inefficiencies of isolated multi-agent on-policy optimization. HACRL enables collaborative optimization with independent execution: heterogeneous agents share verified rollouts during training to mutually improve, while operating independently at inference time. Unlike LLM-based multi-agent reinforcement learning (MARL), HACRL does not require coordinated deployment, and unlike on-/off-policy distillation, it enables bidirectional mutual learning among heterogeneous agents rather than one-directional homogeneous teacher-to-student transfer. Building on this problem, we propose HACPO, a collaborative RL algorithm that enables principled rollout sharing to maximize sample utilization and cross-agent knowledge transfer. To mitigate capability discrepancies and policy distribution shifts, HACPO introduces four tailored mechanisms with theoretical guarantees on unbiased advantage estimation. Extensive experiments across diverse heterogeneous model combinations and reasoning benchmarks show that HACPO consistently improves all participating agents, outperforming GSPO with double rollouts by an average of 3.6% while using only half the rollout cost.

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。

↑