Uncertainty Quantification for Flow-Based Vision-Language-Action Models

2026-06-16

AI Summary

This work addresses the challenge of quantifying predictive uncertainty in existing flow-matching-based vision-language-action models, which often fail to provide reliable failure warnings in out-of-distribution scenarios. To this end, we introduce Velocity Field Divergence (VFD) as a novel measure of epistemic uncertainty tailored for flow-matching models and develop a lightweight ensemble framework for its efficient computation. Building upon VFD, we propose the SAVE framework, which leverages uncertainty estimates to guide active multi-task fine-tuning. Experiments on the LIBERO benchmark demonstrate that VFD effectively identifies task failures with higher accuracy, enabling SAVE to reduce the required number of expert demonstration samples by at least 22% compared to baseline approaches.

Abstract

Vision-language-action models (VLAs) combine vision-language backbones with expressive generative action heads trained via flow matching on large-scale robotic datasets. Despite their strong empirical performance in robotic manipulation, VLAs lack mechanisms to quantify confidence in their predictions and to detect when their actions may be unreliable. This presents a critical limitation for real-world deployment in non-stationary environments, where models inevitably encounter scenarios outside their pretraining distribution and may fail without warning. To address this, we derive an efficient method for quantifying epistemic uncertainty in flow-matching models by leveraging velocity-field disagreement (VFD) across a small ensemble. We successfully use this uncertainty estimate for failure detection during deployment and active fine-tuning of flow-based VLAs. To this end, we propose SAVE, a framework for uncertainty-guided active multitask fine-tuning that reduces the number of costly expert demonstrations required to adapt VLAs to new tasks. Through extensive experiments on the LIBERO benchmark, we demonstrate that VFD yields better-calibrated uncertainty estimates predictive of downstream performance, that VFD achieves strong performance in detecting failures, and that uncertainty-guided data acquisition with SAVE requires at least 22% fewer samples than baselines. In summary, our work shows that quantifying epistemic uncertainty in flow-based VLAs improves both failure awareness and adaptation. Project website: tum-lsy.github.io/uq_vla/.

Problem

• uncertainty quantification
• vision-language-action models
• epistemic uncertainty
• flow matching
• failure detection

Innovation

• Uncertainty Quantification
• Flow Matching
• Velocity-Field Disagreement
• Active Fine-tuning
• Vision-Language-Action Models