Test-Time Perturbation Learning with Delayed Feedback for Vision-Language-Action Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zang, Zehua, Wang, Xi, Sun, Fuchun, Xu, Xiao, Lium, Lixiang, Zhou, Jiahuan, Li, Jiangmeng
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908979986169856
author Zang, Zehua
Wang, Xi
Sun, Fuchun
Xu, Xiao
Lium, Lixiang
Zhou, Jiahuan
Li, Jiangmeng
author_facet Zang, Zehua
Wang, Xi
Sun, Fuchun
Xu, Xiao
Lium, Lixiang
Zhou, Jiahuan
Li, Jiangmeng
contents Vision-Language-Action models (VLAs) achieve remarkable performance in sequential decision-making but remain fragile to subtle environmental shifts, such as small changes in object pose. We attribute this brittleness to trajectory overfitting, where VLAs over-attend to the spurious correlation between actions and entities, then reproduce memorized action patterns. We propose Perturbation learning with Delayed Feedback (PDF), a verifier-free test-time adaptation framework that improves decision performance without fine-tuning the base model. PDF mitigates the spurious correlation through uncertainty-based data augmentation and action voting, while an adaptive scheduler allocates augmentation budgets to balance performance and efficiency. To further improve stability, PDF learns a lightweight perturbation module that retrospectively adjusts action logits guided by delayed feedback, correcting overconfidence issue. Experiments on LIBERO (+7.4\% success rate) and Atari (+10.3 human normalized score) demonstrate consistent gains of PDF in task success over vanilla VLA and VLA with test-time adaptation, establishing a practical path toward reliable test-time adaptation in multimodal decision-making agents. The code is available at \href{https://github.com/zhoujiahuan1991/CVPR2026-PDF}{https://github.com/zhoujiahuan1991/CVPR2026-PDF}.
format Preprint
id arxiv_https___arxiv_org_abs_2604_18107
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Test-Time Perturbation Learning with Delayed Feedback for Vision-Language-Action Models
Zang, Zehua
Wang, Xi
Sun, Fuchun
Xu, Xiao
Lium, Lixiang
Zhou, Jiahuan
Li, Jiangmeng
Computer Vision and Pattern Recognition
Vision-Language-Action models (VLAs) achieve remarkable performance in sequential decision-making but remain fragile to subtle environmental shifts, such as small changes in object pose. We attribute this brittleness to trajectory overfitting, where VLAs over-attend to the spurious correlation between actions and entities, then reproduce memorized action patterns. We propose Perturbation learning with Delayed Feedback (PDF), a verifier-free test-time adaptation framework that improves decision performance without fine-tuning the base model. PDF mitigates the spurious correlation through uncertainty-based data augmentation and action voting, while an adaptive scheduler allocates augmentation budgets to balance performance and efficiency. To further improve stability, PDF learns a lightweight perturbation module that retrospectively adjusts action logits guided by delayed feedback, correcting overconfidence issue. Experiments on LIBERO (+7.4\% success rate) and Atari (+10.3 human normalized score) demonstrate consistent gains of PDF in task success over vanilla VLA and VLA with test-time adaptation, establishing a practical path toward reliable test-time adaptation in multimodal decision-making agents. The code is available at \href{https://github.com/zhoujiahuan1991/CVPR2026-PDF}{https://github.com/zhoujiahuan1991/CVPR2026-PDF}.
title Test-Time Perturbation Learning with Delayed Feedback for Vision-Language-Action Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.18107