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

Ralf Römer, Maximilian Seeliger, Saida Liu, Ben Sturgis, Marco Bagatella, Daniel Marta, Andreas Krause, Angela P. Schoellig · ETH Zürich / University of Toronto (Learning Systems & Robotics Lab) · 2026

Summary

Quantifies epistemic uncertainty in flow-matching VLAs using velocity-field disagreement (VFD) across a small ensemble, enabling failure detection at deployment and sample-efficient active fine-tuning (SAVE).

Abstract Summary

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, the authors derive an efficient method for quantifying epistemic uncertainty in flow-matching models by leveraging velocity-field disagreement (VFD) across a small ensemble. They use this uncertainty estimate for failure detection during deployment and active fine-tuning of flow-based VLAs. They 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, they 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.

Key Points

Introduces velocity-field disagreement (VFD) for epistemic uncertainty quantification in flow-matching VLAs
Uses uncertainty for failure detection during deployment
Proposes SAVE: uncertainty-guided active multitask fine-tuning requiring ≥22% fewer expert demonstrations
Validated on LIBERO benchmark with better-calibrated uncertainty than baselines
RSS 2026 Best Paper Award

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

Authors: Ralf Römer, Maximilian Seeliger, Saida Liu, Ben Sturgis, Marco Bagatella, Daniel Marta, Andreas Krause, Angela P. Schoellig

Venue: RSS 2026 — Best Paper Award | Year: 2026

arXiv: 2606.18043 | PDF: link

Summary

Vision-language-action models (VLAs) combine vision-language backbones with expressive generative action heads trained via flow matching. While powerful, they lack any mechanism to quantify confidence or detect unreliable predictions — a critical gap for real-world deployment where out-of-distribution scenarios cause silent failures.

This paper introduces velocity-field disagreement (VFD), an efficient epistemic uncertainty estimate for flow-matching models. By measuring disagreement across a small ensemble of velocity fields, VFD provides calibrated uncertainty without significant computational overhead.

Key Contributions

VFD — Velocity-Field Disagreement: A scalable epistemic uncertainty quantification method specifically designed for flow-matching action heads in VLAs.
Failure Detection: VFD uncertainty scores effectively flag unreliable actions at deployment time, enabling safe intervention or fallback.
SAVE — Sample-efficient Active VLA finE-tuning: An uncertainty-guided active learning framework that selects which demonstrations to collect, reducing expert demonstration requirements by at least 22% compared to baselines on the LIBERO benchmark.

Why It Matters

Safety: Flow-based VLAs can now express “I don’t know” — essential for non-stationary real-world deployment.
Data Efficiency: Active fine-tuning with SAVE directly reduces the cost of adapting VLAs to new tasks.
Award: RSS 2026 Best Paper — signals high-impact recognition from the robotics community.

Code & Data

Resource	Link
Project page	https://tum-lsy.github.io/uq_vla/
Code	https://github.com/learnsyslab/uq_vla
arXiv	https://arxiv.org/abs/2606.18043
Benchmark	LIBERO

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

Summary

Abstract Summary

Key Points

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

Summary

Key Contributions

Why It Matters

Code & Data

Related Papers

DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation

GuidedVLA: Specifying Task-Relevant Factors via Plug-and-Play Action Attention Specialization

From Imagined Futures to Executable Actions: Mixture of Latent Actions for Robot Manipulation

Pre-VLA: Preemptive Runtime Verification for Reliable Vision-Language-Action and World-Model Rollouts