---
title: "STAR: Steering Action Representations during Inference"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.41009v1"
permalink: "https://feed.astralor.com/items/itm_404d80786b9feec368c1d9ee"
published: "2026-10-29T21:42:17.000Z"
score: 53
channel: "tech"
tags: ["推理效率","RAG","多模态","STAR","Steering Action Representations","Inference Vision-language-action","VLAs","However"]
---

# STAR: Steering Action Representations during Inference

## At a glance

Vision-language-action models (VLAs) map language instructions and visual observations to robot actions, showing strong capabilities across robotic manipulation tasks. However, in our probing experiment, we find that ac…

## Key points

- Vision-language-action models (VLAs) map language instructions and visual observations to robot actions, showing strong capabilities across robotic manipulation tasks.
- However, in our probing experiment, we find that action representations near the action output do not fully preserve task-relevant visual grounding.
- Motivated by this late-layer grounding gap, we introduce STAR, for STeering Action Representations, a training-free inference method for pre-trained VLAs.

## Full text

Vision-language-action models (VLAs) map language instructions and visual observations to robot actions, showing strong capabilities across robotic manipulation tasks. However, in our probing experiment, we find that action representations near the action output do not fully preserve task-relevant visual grounding. 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. Experiments on robot manipulation benchmarks show that STAR improves late-layer object awareness and increases average task success rates by 4.0 percentage points. Our code is available at https://github.com/PKU-VaLuE-Lab/STAR.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_404d80786b9feec368c1d9ee