arXiv · cs.AI · Guangyu Zhao, Yongchang Zhang, Haoxuan Ru, Kaichen He, Haoran Wen, Kun Zhan, Zeyu Zhang, Yin Li, Yiwu Zhong · 9 小时前模型评分36
STAR：在推理阶段引导视觉-语言-动作模型的动作表征
STAR: Steering Action Representations during Inference
导读
针对视觉-语言-动作模型（VLA，Vision-Language-Action）动作输出附近的表征未能充分保留任务相关视觉定位的问题，研究者提出免训练推理方法 STAR，对后层残差更新做指数移动平均，让后层动作表征继承前层的物体感知更新方向，并用本体状态调节混合强度。在机器人操作基准上，STAR 将平均任务成功率提升 4.0 个百分点，代码已公开。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “Motivated by this late-layer grounding gap, we introduce STAR, for STeering Action Representations, a training-free inference method for pre-trained VLAs.”
- “STAR applies an exponential moving average over late-layer residual updates, allowing later action representations to inherit object-aware update directions from earlier layers without model updates or architectural changes.”
- “It further uses recent proprioceptive states to adapt the mixing strength, using conservative mixing when execution dynamics change rapidly.”
- “Our code is available at this https URL .”
来源：arXiv · cs.AI · arxiv.org