从像素到动作：原生视觉-语言-动作策略 NEO-VLA
原标题：Paper page - From Pixels to Actions - Towards Native Vision-Language-Action Policies
AI 摘要
该论文提出 NEO-VLA，一个结构上无编码器的视觉-语言-动作（VLA）框架：图像由轻量 patch 卷积层 token 化，并与语言在主干网络中共同演化，使 flow-matching 动作专家能够从不同深度的指令条件视觉表示中提取信息。受控对比显示，原生架构优于基于编码器的方案，全量微调可缩小但无法消除差距，单纯扩大规模也无法弥补；逐层探测表明动作关键信号常在最终视觉表示之前达到峰值。在多种操作场景和分布偏移下，NEO-VLA 实现了更强、更鲁棒的控制。
正文节选
From Pixels to Actions - Towards Native Vision-Language-Action Policies Abstract NEO-VLA eliminates fixed visual encoders to let action policies access evolving visual representations across network depths, improving manipulation robustness. Vision-language-action models are typically built on a simple premise: perception is first compressed by visual encoders, and action reasoning begins only afterwards. Yet manipulation often hinges on precisely the visual signals most vulnerable to this abstr