外观
论文arXiv 2610.0631823 天前
接线很重要：VLA 策略中可供性头的注入拓扑与初始化
原文链接：Wiring Matters: Injection Topology and Initialization of Affordance Heads in Vision-Language-Action Policies（arXiv 2610.06318 · 发布日期：2026-10-05）
Wiring Matters: Injection Topology and Initialization of Affordance Heads in Vision-Language-Action Policies
研究VLA中可供性头损害指令遵循的问题，提出stop-gradient读取与桥接重注入，并在LIBERO验证性能。
核心要点
- 密集可供性监督虽有吸引力，但朴素共训可供性头会严重损害指令遵循。
- 方法以stop-gradient读取骨干，并通过学习桥将中间头特征重注入动作专家。
- 在LIBERO上，允许梯度回传使策略低于无头基线，保护骨干是前提。
- 初始化比注入拓扑更关键；主动初始化残差桥达96.2%，额外参数低于1%。
验证范围与局限：验证范围为LIBERO benchmark上的现代VLA，摘要未说明真机实验。
英文摘要 (Abstract)
Dense affordance supervision is an appealing auxiliary signal for vision-language-action (VLA) policies, yet naively co-training an affordance head can severely damage instruction following. We present a controlled study of how to wire such a head into a modern VLA on the LIBERO benchmark. Our recipe reads the backbone through a stop-gradient and re-injects an intermediate head feature into the action expert via a learned bridge. The stop-gradient is a precondition: letting affordance gradients reach the backbone drops the policy below the headless base (85.5% vs. 93.1%). With the backbone protected, a same-budget 2*2 ablation over injection topology (concatenation vs. residual) and bridge initialization (zero vs. random) shows initialization is the dominant lever. The best wiring, an actively initialized residual bridge, reaches 96.2%, matching the far more elaborate three-expert AffordanceVLA (95.8%) with under 1% extra parameters. Two probes explain the mechanism: ground-truth affordances fed as an input hurt, and inference-time zeroing shows a lazy bridge acts only as a training-time regularizer while an active bridge becomes load-bearing.
作者：Zijian An, Linhan Wang, Jiayan Wang, Shijie Geng, Ran Yang, Yiming Feng, Lifeng Zhou
来源
本页事实对应的公开页面，以原文为准。