English(EN) When to Stop Penalizing Actions? Martingale GRPO for VLA Reinforcement Learning

Martingale GRPO 通过优化动作惩罚来增强视觉语言模型强化学习

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-26 04:00

研究人员开发了 Martingale GRPO (MART-GRPO)，这是一种新颖的强化学习技术，旨在提高视觉语言模型 (VLM) 训练的效率。与将单一优势分数分配给轨迹中所有动作的传统 GRPO 方法不同，MART-GRPO 通过识别失败反馈的最佳停止点来选择性地惩罚动作。这种方法可以防止有用的恢复动作与错误一起被惩罚。在 RoboTwin 2.0 平台上使用 OpenVLA-OFT 模型进行的实验表明，与标准 GRPO 相比，MART-GRPO 显著提高了成功率。 AI

影响这项研究可能导致更有效的视觉语言模型训练，从而可能提高它们在现实世界机器人应用中的性能。

排序理由该集群包含一篇详细介绍强化学习新算法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 14 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍强化学习新算法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Zongyao Zhao, Cheng-Jie Zhao, Ruihang Jiang, Bing Zheng, Zhaolin Wang, Yuanwei Liu · 2026-10-26 04:00

何时停止惩罚行为？用于VLA强化学习的Martingale GRPO

arXiv:2610.32431v1 Announce Type: new Abstract: A failed vision-language-action (VLA) rollout does not imply that every action contributed to failure. Yet in group-relative policy optimization (GRPO), a single trajectory-level advantage is assigned to all actions in a rollout. Po…

本文如何被排名

报道来源 [1]

何时停止惩罚行为？用于VLA强化学习的Martingale GRPO

相关实体

相关话题