arXiv 2509.226432026-03-01cs.RO

VLA-Reasoner: 通过在线蒙特卡洛树搜索增强视觉-语言-动作模型的推理能力

VLA-Reasoner: Empowering Vision-Language-Action Models with Reasoning via Online Monte Carlo Tree Search

School of Electrical and Electronic Engineering, Nanyang Technological University（南洋理工大学电子与电气工程学院）；Tsinghua Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生院）；School of Intelligent Engineering and Automation, Beijing University of Posts and Telecommunications（北京邮电大学智能工程与自动化学院）

Wenkai Guo, Guanxing Lu, Haoyuan Deng, Zhenyu Wu, Yansong Tang, Ziwei Wang

备注：: 8 pages, 6 figures, Accepted by ICRA 2026

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

VLA-Reasoner通过在线蒙特卡洛树搜索增强视觉-语言-动作模型，提升长时间轨迹任务的推理能力与执行效率。

摘要：

视觉-语言-动作模型（VLAs）通过扩展模仿学习在一般机器人操作任务中实现了强大的性能。然而，现有的VLAs仅能预测短视的下一步动作，由于累积偏差，难以处理长时间轨迹任务。为了解决这个问题，我们提出了一种名为\method的插件框架，该框架有效增强了现成的VLAs，使其能够通过测试时间扩展预测未来状态。具体而言，\method采样并展开可能的动作轨迹，其中涉及的动作是推理，通过世界模型生成未来状态，使\method能够预见并推理潜在结果，搜索最佳动作。我们进一步利用蒙特卡洛树搜索（MCTS）来提高在大动作空间中的搜索效率，其中逐步的VLA预测作为根节点。同时，我们引入基于核密度估计（KDE）的置信度采样机制，以在MCTS中实现高效的探索，而无需冗余的VLA查询。我们通过离线价值估计策略评估MCTS中的中间状态，以评分预测的未来并利用长期反馈纠正偏差。我们在模拟器和现实世界中进行了广泛的实验，证明了所提出的VLA-Reasoner在现有VLAs上实现了显著的改进。我们的方法展示了通往可扩展测试时间计算的机器人操作的潜在途径。项目网站可在：https://vla-reasoner.github.io/上找到。

英文摘要：

Vision-Language-Action models (VLAs) achieve strong performance in general robotic manipulation tasks by scaling imitation learning. However, existing VLAs are limited to predicting short-sighted next-action, which struggle with long-horizon trajectory tasks due to incremental deviations. To address this problem, we propose a plug-in framework named \method that effectively empowers off-the-shelf VLAs with the capability of foreseeing future states via test-time scaling. Specifically, \method samples and rolls out possible action trajectories where involved actions are rationales to generate future states via a world model, which enables \method to foresee and reason potential outcomes and search for the optimal actions. We further leverage Monte Carlo Tree Search (MCTS) to improve search efficiency in large action spaces, where step-wise VLA predictions seed the root. Meanwhile, we introduce a confidence sampling mechanism based on Kernel Density Estimation (KDE), to enable efficient exploration in MCTS without redundant VLA queries. We evaluate intermediate states in MCTS via an offline value estimation strategy, to score predicted futures and correct deviations with long-term feedback. We conducted extensive experiments in both simulators and the real world, demonstrating that our proposed VLA-Reasoner achieves significant improvements over the state-of-the-art VLAs. Our method highlights a potential pathway toward scalable test-time computation of robotic manipulation. The project website is available at: https://vla-reasoner.github.io/.

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。