arXiv cs.AI· Yuan Xu, Youheng Shi, Chengyang Li, Wentao Zhu, Yizhou Wang·· 12 小时前AI 评分38
机器人动作表示的语义锚定方法提升 VLA 模型泛化能力
Semantic Anchoring for Robotic Action Representations
AI 导读
研究提出一种即插即用的语义锚定方法，将 VLA 模型的动作表示锚定到语义流形，并拆分为共享语义通道与私有通道，推理时全部丢弃、不改变部署模型。该方法在多个 VLA 骨干和仿真、真实基准上验证，真实分布内任务最高提升 18.7%，分布外泛化最高提升 21.5%。系统探测还发现，预训练编码器的语义结构会在微调中退化，其质量与任务成功率和分布外泛化同步。
正文 · 原文
Abstract:Vision-Language-Action (VLA) models inherit rich semantic representations from pretrained Vision-Language Models, yet fine-tuning on limited robot demonstrations degrades this structure and undermines generalization. A fundamental question therefore arises: what constitutes a good action representation? Inspired by the mirror neuron theory's insight that observation and execution share an intention-level encoding, we examine whether a robot's action representations preserve the semantic structure captured by pretrained encoders. Systematic probing confirms that this structure erodes during finetuning, and that its quality synchronizes with both task success and out-of-distribution generalization. We further introduce a plug-and-play method that anchors action representations to a semantic manifold while decomposing representations into a shared semantic channel and a private channel, all discarded at inference, leaving the deployed model unchanged. Validated on different VLA backbones across simulation and real-world benchmarks, our method yields up to +18.7% on real-world in-distribution tasks and +21.5% on out-of-distribution generalization.
| Comments: | Accepted at CoRL 2026. Project page: this https URL |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2607.13597 [cs.RO] |
| (or arXiv:2607.13597v3 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2607.13597 arXiv-issued DOI via DataCite |
Submission history
From: Yuan Xu [view email]
[v1]
Wed, 15 Jul 2026 08:45:15 UTC (3,863 KB)
[v2]
Sat, 18 Jul 2026 04:17:04 UTC (3,863 KB)
[v3]
Fri, 23 Oct 2026 17:10:14 UTC (4,034 KB)
来源：arXiv cs.AI · arxiv.org