aiqbench analysis
This research addresses a critical gap in generalist robot policies by introducing an efficient method to quantify epistemic uncertainty using velocity-field disagreement in flow-matching models. By leveraging this metric, the authors enable reliable failure detection and propose SAVE, an active fine-tuning strategy that significantly reduces the need for costly expert demonstrations while improving task success rates. The work advances the practical deployment of VLAs and WAMs by ensuring they can operate safely in non-stationary environments where out-of-distribution scenarios are common.
Uncertainty Quantification for Flow-Based Generalist Robot Policies
arXiv:2606.18043v2 Announce Type: replace-cross Abstract: Generalist robot policies, such as vision-language-action models (VLAs) and world-action models (WAMs), combine powerful pretrained backbones with expressive generative action heads trained via flow matching on large-scale robotic datasets. Despite their strong empirical performance in robotic manipulation, these policies 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 to quantify epistemic uncertainty in flow-matching models by leveraging velocity-field disagreement (VFD) across a small ensemble. We successfully use this uncertainty estimate for detecting failures during deployment and active fine-tuning of flow-based generalist policies. For the latter, we propose SAVE, a simple yet effective method for uncertainty-guided active multitask fine-tuning that reduces the number of costly expert demonstrations required to adapt generalist policies to new tasks. We conduct experiments in simulation and the real world, across VLAs and a WAM. VFD yields better-calibrated uncertainty estimates predictive of downstream performance and detects failures with 8 pp higher overall accuracy than existing methods. Across three real-world tasks, SAVE improves final average success from 39 % to 47 % with a fixed demonstration budget. Our results show that measuring epistemic uncertainty with VFD enhances both failure awareness and adaptation of generalist robot policies. Project website: tum-lsy.github.io/uq_generalist_policies.
Abstract:Generalist robot policies, such as vision-language-action models (VLAs) and world-action models (WAMs), combine powerful pretrained backbones with expressive generative action heads trained via flow matching on large-scale robotic datasets. Despite their strong empirical performance in robotic manipulation, these policies 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 to quantify epistemic uncertainty in flow-matching models by leveraging velocity-field disagreement (VFD) across a small ensemble. We successfully use this uncertainty estimate for detecting failures during deployment and active fine-tuning of flow-based generalist policies. For the latter, we propose SAVE, a simple yet effective method for uncertainty-guided active multitask fine-tuning that reduces the number of costly expert demonstrations required to adapt generalist policies to new tasks. We conduct experiments in simulation and the real world, across VLAs and a WAM. VFD yields better-calibrated uncertainty estimates predictive of downstream performance and detects failures with 8 pp higher overall accuracy than existing methods. Across three real-world tasks, SAVE improves final average success from 39 % to 47 % with a fixed demonstration budget. Our results show that measuring epistemic uncertainty with VFD enhances both failure awareness and adaptation of generalist robot policies. Project website: this http URL.
Project page: this http URL. 41 pages, 18 figures
Robotics (cs.RO); Machine Learning (cs.LG)
I.2.6; I.2.9; I.2.10
arXiv:2606.18043 [cs.RO]
(or arXiv:2606.18043v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.18043 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 340 words
Related coverage
Perturbation-Based Epistemic Uncertainty for Failure Detection in Vision-Language-Action Models
arXiv · Sep 2, 2026
2026 BAIR Graduate Showcase
BAIR Blog · Jul 1, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026
Uncertainty Quantification for Flow-Based Vision-Language-Action Models
arXiv · Jun 17, 2026