aiqbench analysis
This research introduces GC-VLA, a vision-language-action framework that leverages geometric supervision of object changes to enable embodiment-agnostic pretraining across robot and egocentric video datasets. By decoupling the prediction of geometry changes from specific action flows, the model achieves near-perfect success rates on manipulation benchmarks while maintaining inference efficiency through frozen components. The approach represents a significant step toward generalizable robotic policies that can transfer knowledge between different physical platforms without extensive retraining.
HABILIS Brain 0: Geometry-Change Supervision for Vision-Language-Action and Residual Flow Recovery
arXiv:2609.25558v1 Announce Type: new Abstract: Vision-language-action policies benefit from geometric supervision, but current-frame geometry alone does not explicitly describe the changes associated with manipulation. This design is motivated by the goal of learning an embodiment-agnostic visual interface that can be pretrained across robot and egocentric video before robot-specific action alignment. We introduce Geometry-Change VLA (GC-VLA), which learns to predict multiview future-current geometry-change tokens from current observations. Offline frame pairs define a nominal 0.5-second prediction horizon; future observations are used only to construct training targets. Stage 1 trains a geometry-change vision-language model (GC-VLM). Stage 2 introduces a continuous ActionExpert and aligns it with robot actions while stopping action-flow gradients at the VLM interface. Stage 3 enables these gradients to update the trainable VLM components jointly with the ActionExpert. Stage 4 freezes GC-VLA and applies Geometry-Conditioned Residual Flow (GCRF), using a binary intervention router and a single bounded residual velocity policy learned from closed-loop feedback. GC-VLA achieves 95.20% success on LIBERO, and GC-VLA with GCRF achieves 99.55%. Inference uses current observations and the learned GC representation without executing the offline target encoders.
Abstract:Vision-language-action policies benefit from geometric supervision, but current-frame geometry alone does not explicitly describe the changes associated with manipulation. This design is motivated by the goal of learning an embodiment-agnostic visual interface that can be pretrained across robot and egocentric video before robot-specific action alignment. We introduce Geometry-Change VLA (GC-VLA), which learns to predict multiview future-current geometry-change tokens from current observations. Offline frame pairs define a nominal 0.5-second prediction horizon; future observations are used only to construct training targets. Stage 1 trains a geometry-change vision-language model (GC-VLM). Stage 2 introduces a continuous ActionExpert and aligns it with robot actions while stopping action-flow gradients at the VLM interface. Stage 3 enables these gradients to update the trainable VLM components jointly with the ActionExpert. Stage 4 freezes GC-VLA and applies Geometry-Conditioned Residual Flow (GCRF), using a binary intervention router and a single bounded residual velocity policy learned from closed-loop feedback. GC-VLA achieves 95.20% success on LIBERO, and GC-VLA with GCRF achieves 99.55%. Inference uses current observations and the learned GC representation without executing the offline target encoders.
Robotics (cs.RO); Machine Learning (cs.LG)
arXiv:2609.25558 [cs.RO]
(or arXiv:2609.25558v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.25558 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 · 271 words
Related coverage
Track4Action: Distilling World-Centric 3D Tracker into Vision-Language-Action Policies
arXiv · Aug 5, 2026
TemporalFlow-VLA: Learning Physically Grounded Execution History for Long-Horizon Robot Manipulation
arXiv · Aug 28, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026
GALA: Geometry-Aware Latent Action Modeling for Vision-Language-Action Model Pretraining across Embodiments
arXiv · Sep 21, 2026