arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2507.23391cs.LGcs.RO
收藏

无需奖励建模的从大视觉语言模型反馈中进行策略学习

Policy Learning from Large Vision-Language Model Feedback without Reward Modeling

School of Electrical Engineering, KAIST(韩国科学技术院电子工程学院)

Tung M. Luu, Donghoon Lee, Younghwan Lee, Chang D. Yoo

AI总结:

针对离线强化学习需人工设计奖励函数的瓶颈,提出PLARE方法,直接利用VLM的轨迹偏好标签通过有监督对比偏好学习训练策略,无需显式奖励建模,在仿真与现实机器人操作任务上性能优异。

摘要:

离线强化学习(RL)提供了一个强大的框架,可利用预先收集的次优数据集训练机器人智能体,无需进行成本高昂、耗时且可能存在危险的在线交互。这在安全关键型现实世界应用中尤为有用,这类场景下的在线数据收集成本高昂且不切实际。然而,现有的离线RL算法通常需要带奖励标签的数据,这带来了额外的瓶颈:奖励函数设计本身成本高、劳动密集,且需要大量领域专业知识。本文提出了PLARE,这是一种利用大视觉语言模型(VLM)为智能体训练提供引导信号的新方法。PLARE无需依赖人工设计的奖励函数,而是根据语言任务描述,向VLM查询成对视觉轨迹片段的偏好标签。随后,策略通过有监督对比偏好学习目标直接从这些偏好标签中训练,无需学习显式奖励模型。通过在MetaWorld的机器人操作任务上开展的大量实验,PLARE的性能与现有最先进的基于VLM的奖励生成方法相当甚至更优。此外,我们在实体机器人的现实世界操作任务中验证了PLARE的有效性,进一步证实了其实际适用性。

英文摘要:

Offline reinforcement learning (RL) provides a powerful framework for training robotic agents using pre-collected, suboptimal datasets, eliminating the need for costly, time-consuming, and potentially hazardous online interactions. This is particularly useful in safety-critical real-world applications, where online data collection is expensive and impractical. However, existing offline RL algorithms typically require reward labeled data, which introduces an additional bottleneck: reward function design is itself costly, labor-intensive, and requires significant domain expertise. In this paper, we introduce PLARE, a novel approach that leverages large vision-language models (VLMs) to provide guidance signals for agent training. Instead of relying on manually designed reward functions, PLARE queries a VLM for preference labels on pairs of visual trajectory segments based on a language task description. The policy is then trained directly from these preference labels using a supervised contrastive preference learning objective, bypassing the need to learn explicit reward models. Through extensive experiments on robotic manipulation tasks from the MetaWorld, PLARE achieves performance on par with or surpassing existing state-of-the-art VLM-based reward generation methods. Furthermore, we demonstrate the effectiveness of PLARE in real-world manipulation tasks with a physical robot, further validating its practical applicability.

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。

↑