aiqbench analysis
This research systematically evaluates how Vision-Language-Action (VLA) models should integrate proprioceptive state inputs, addressing critical questions about timing, history length, and architectural placement for closed-loop control. By testing five distinct interfaces across 45 tasks with varying temporal histories, the study establishes testable design principles that clarify whether current frame data or historical sequences genuinely enhance performance beyond mere conditioning capacity. These findings provide essential guidance for optimizing robot autonomy architectures where precise state representation is central to execution fidelity.
How Should Vision-Language-Action Models Use Proprioceptive State?
arXiv:2608.03052v1 Announce Type: new Abstract: Recent Vision-Language-Action (VLA) models almost universally take robot proprioceptive state as input, yet wire it in incompatible ways -- serialized into text prompts, projected into the vision-language prefix, or fed directly to the action expert -- and almost always as a single current frame. Three questions remain open: (1) whether, and on which tasks, current state actually improves closed-loop control; (2) how much state history helps, and whether its benefit reflects genuine temporal variation rather than added conditioning capacity; and (3) where state should enter the model -- the vision-language backbone or the action-generation module. We answer these questions through controlled experiments on a flow-matching VLA, fixing the backbone, training data, action representation, and evaluation protocol throughout. We implement five representative interfaces -- discrete state prompt, VLM prefix, action prefix, state expert, and feature modulation -- under matched implementation details, and evaluate them on 45 atomic tasks spanning three task families plus 20 composite tasks; we then sweep the state-history length from 1 to 96 frames to examine how historical state information affects model performance. The experiments yield systematic answers to all three questions, distilled into testable design principles for state-aware VLAs.
Title:How Should Vision-Language-Action Models Use Proprioceptive State?
Abstract:Recent Vision-Language-Action (VLA) models almost universally take robot proprioceptive state as input, yet wire it in incompatible ways -- serialized into text prompts, projected into the vision-language prefix, or fed directly to the action expert -- and almost always as a single current frame. Three questions remain open: (1) whether, and on which tasks, current state actually improves closed-loop control; (2) how much state history helps, and whether its benefit reflects genuine temporal variation rather than added conditioning capacity; and (3) where state should enter the model -- the vision-language backbone or the action-generation module. We answer these questions through controlled experiments on a flow-matching VLA, fixing the backbone, training data, action representation, and evaluation protocol throughout. We implement five representative interfaces -- discrete state prompt, VLM prefix, action prefix, state expert, and feature modulation -- under matched implementation details, and evaluate them on 45 atomic tasks spanning three task families plus 20 composite tasks; we then sweep the state-history length from 1 to 96 frames to examine how historical state information affects model performance. The experiments yield systematic answers to all three questions, distilled into testable design principles for state-aware VLAs.
Robotics (cs.RO)
arXiv:2608.03052 [cs.RO]
(or arXiv:2608.03052v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2608.03052 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 292 words
Related coverage
NeuroVLA: A Brain-inspired Embodied Intelligence for Fluid and Fast Reflexive Robotics Control
arXiv · Oct 15, 2026
{\tau}: Learning Touch-Augmented Vision-Language-Action Models from Future Visual Supervision
arXiv · Jul 28, 2026
Towards the Vision-Sound-Language-Action Paradigm: The HEAR Framework for Sound-Centric Manipulation
arXiv · Sep 22, 2026
StepTrigger: Contact-State-Triggered Backdoor Attacks on VLM-Powered Legged Robots
arXiv · Sep 24, 2026