ICRA 2026 | 新加坡国立大学提出FD-VLA：力蒸馏视觉语言动作模型，攻克接触丰富操作
Quick Take
The FD-VLA model, proposed by National University of Singapore, enhances robot manipulation in contact-rich tasks by integrating force signals through a distillation mechanism, improving success rates significantly over traditional methods. This approach reduces reliance on expensive sensors while maintaining effective force representation during operation.
Key Points
- FD-VLA integrates force perception into the Vision-Language-Action model for better manipulation.
- The model predicts force labels without physical sensors, enhancing robustness.
- Experiments show FD-VLA outperforms traditional methods in tasks like button pressing and plug insertion.
- This approach advances embodied intelligence by focusing on contact understanding.
- The research highlights the importance of force distillation in practical VLA applications.
Article Excerpt
From source RSS / original summary论文原文链接arXiv预印本页面：https://arxiv. org/abs/2602. 02142v2arXiv全文PDF：https://arxiv. org/pdf/2602. 02142v2DBLP页面：https://dblp. uni-trier. de/pid/00/9938. html（内含更详细的引用信息）Semantic Scholar页面：https://www. semanticscholar. org/paper/FD-VLA:-Force-Distilled-Vision-Language-Action-for-Zhao-Wang/84496e9c36fa5b863f5702abb1dbc5560ee7db5b 原文作者：公众号“计算机顶会大全”原文链接：https://mp. weixin. qq.
com/s/SbiHonAq0qYEP-sC-sB-bA ICRA 2026| FD-VLA: Force-Distilled Vision-Language-Action Model for Contact-Rich Manipulation过去，VLA模型主要依赖视觉和语言来完成机器人操作，但在真实场景中，很多任务并不是看见就能做好。 比如插头插入、按钮按压、擦白板等接触丰富任务，真正影响成功率的是接触力、阻力变化、摩擦和细微形变。 这篇论文关注的正是VLA的力觉短板。 论文提出FD-VLA，通过力蒸馏机制，让模型在训练阶段利用真实力信号学习潜在力表示；在推理阶段，则不再依赖实体力传感器，而是根据视觉观察和机器人本体状态预测力标记，并与视觉语言信息共同生成动作。 这样既保留了力觉对接触操作的帮助，又降低了真实部署时对昂贵、脆弱传感器的依赖。 创新点主要在于三方面：一是把力觉信息以蒸馏方式融入VLA，而不是简单拼接原始力信号；二是利用视觉和本体状态预测接触相关力表示，提升任务相关性和鲁棒性；三是在真实机器人平台上验证擦白板、按按钮、插头插入等典型接触任务。
实验表明，FD-VLA整体成功率明显高于无力觉版本和直接输入原始力信号的方法，说明可学习的力表示比粗暴使用传感器数据更有效。 这篇论文的价值在于，它把VLA从看懂再行动推进到理解接触再行动。 对具身智能研究来说，力觉蒸馏、触觉增强、状态建模和接触丰富操作，正在成为VLA实用化落地的重要发文切口。 雷峰网
Reader Mode unavailable (could not extract clean content).
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from 雷峰网机器人
See more →
ICRA 2026 | 港中文高源、林天麟团队提出自发共适应策略：基于元学习赋能异构多机器人系统协同进化
The Hong Kong Polytechnic University team presented a meta-learning framework for heterogeneous multi-robot systems at ICRA 2026, enhancing adaptability in dynamic environments. Their approach improved task completion efficiency by 21% and reduced human burden by 39%, demonstrating effective human-robot interaction.
