aiqbench analysis
LARA introduces a plug-and-play framework that jointly optimizes Latent Action Models and Vision-Language-Action models through representation alignment, directly addressing the data scarcity and training decoupling issues in robotic manipulation. By enabling reciprocal benefits where latent models avoid spurious visual changes while VLAs are regularized against hallucinations, the approach yields significant performance gains across both simulation and real-world benchmarks. This method offers a practical pathway for enhancing VLA capabilities without requiring extensive new real-world datasets.
LARA: Latent Action Representation Alignment for Vision-Language-Action Models
arXiv:2606.07100v1 Announce Type: cross Abstract: Visual-language action (VLA) models enable robots to predict actions directly from observations and language instructions, but their performance depends on large-scale, high-quality data and is limited by the scarcity of real-world robot action datasets. To facilitate VLA model learning with abundant unlabeled human videos, Latent Action Models (LAM) learn latent action representations from visual dynamics to provide additional supervision for VLA learning. However, LAM and VLA are typically trained separately, leaving LAM ungrounded during VLA training and VLA models constrained by frozen LAM representations. To address these issues, we propose Latent Action Representation Alignment (LARA), a plug-and-play framework that jointly optimizes LAM and VLA via representation alignment. This enables reciprocal benefits where LAMs learn with action trajectories to avoid spurious visual changes, while VLAs are regularized by forward dynamics learned within LAMs to reduce hallucinations of functionally ineffective trajectories. We demonstrate LARA versatility and effectiveness for pre-training, post-training enhancement of pre-trained VLA models, and LAM refinement, achieving an average of ~10%, ~5%, and ~15% improvement over 3 simulation and 1 meticulously designed real-world robotic manipulation benchmarks.
Abstract:Visual-language action (VLA) models enable robots to predict actions directly from observations and language instructions, but their performance depends on large-scale, high-quality data and is limited by the scarcity of real-world robot action datasets. To facilitate VLA model learning with abundant unlabeled human videos, Latent Action Models (LAM) learn latent action representations from visual dynamics to provide additional supervision for VLA learning. However, LAM and VLA are typically trained separately, leaving LAM ungrounded during VLA training and VLA models constrained by frozen LAM representations. To address these issues, we propose Latent Action Representation Alignment (LARA), a plug-and-play framework that jointly optimizes LAM and VLA via representation alignment. This enables reciprocal benefits where LAMs learn with action trajectories to avoid spurious visual changes, while VLAs are regularized by forward dynamics learned within LAMs to reduce hallucinations of functionally ineffective trajectories. We demonstrate LARA versatility and effectiveness for pre-training, post-training enhancement of pre-trained VLA models, and LAM refinement, achieving an average of ~10%, ~5%, and ~15% improvement over 3 simulation and 1 meticulously designed real-world robotic manipulation benchmarks.
Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
arXiv:2606.07100 [cs.CV]
(or arXiv:2606.07100v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2606.07100 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 · 275 words
Related coverage
HARP-VLA: Human-Robot Aligned Representation Learning for Vision-Language-Action Model
arXiv · Oct 12, 2026
VLAFlow: A Unified Training Framework for Vision-Language-Action Models via Co-training and Future Latent Alignment
arXiv · Aug 5, 2026
PAIR: Bridging Perception and Action in Vision-Language-Action Models
arXiv · Oct 8, 2026
PHR-VLA: Planning Horizon Reasoning for Vision-Language-Action Models
arXiv · Sep 1, 2026