---
title: "CalVLA: Shaping Native Confidence in Vision-Language-Action Policies with Outcome Supervision"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2611.00394v1"
permalink: "https://feed.astralor.com/items/itm_840cb66dffac559a0ca9e3c9"
published: "2026-10-30T21:18:15.000Z"
score: 54
channel: "ai"
tags: ["安全","RAG","多模态","CalVLA","Shaping Native Confidence","Vision-Language-Action Policies","Outcome Supervision Can","VLA"]
---

# CalVLA: Shaping Native Confidence in Vision-Language-Action Policies with Outcome Supervision

## 速览

Can a vision-language-action (VLA) policy signal likely task failure from its own action probabilities? We introduce CalVLA, which shapes native action-token confidence using successful and failed rollouts and an Action…

## 关键点

- Can a vision-language-action (VLA) policy signal likely task failure from its own action probabilities?
- We introduce CalVLA, which shapes native action-token confidence using successful and failed rollouts and an Action-Confidence Alignment (ACal) squared margin-ranking loss.
- Low-rank adapters update OpenVLA-7B without an auxiliary failure detector.

## 全文

Can a vision-language-action (VLA) policy signal likely task failure from its own action probabilities? We introduce CalVLA, which shapes native action-token confidence using successful and failed rollouts and an Action-Confidence Alignment (ACal) squared margin-ranking loss. Low-rank adapters update OpenVLA-7B without an auxiliary failure detector. Observations are paired within tasks at similar normalized elapsed times, not necessarily equivalent physical progress. On a held-out test split from 500 LIBERO-10 configurations, adapted confidence achieves a joint failure-discrimination AUROC of $0.7392\pm0.0480$ over three seeds, averaging scores at 130 and 260 steps. Frozen-policy probes reach 0.6146--0.6436, but superiority over the linear probe is statistically inconclusive. For the same confidence summary, AUROC rises from 0.4485 before adaptation to 0.7327 after adaptation. Mean task success decreases from 58% to 55%. These results show improved failure discrimination, not improved control or established safety benefits.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_840cb66dffac559a0ca9e3c9