aiqbench analysis
GeoAlign advances Vision-Language-Action (VLA) models by integrating geometry-aware spatial alignment to improve robotic manipulation beyond simple semantic grounding. The architecture utilizes offline RGB-D supervision to generate compact geometry tokens that guide action prediction using only RGB, language, and proprioceptive state during rollout. This approach demonstrates significant performance gains on both simulated and real-world benchmarks, addressing a critical gap in executable robot control systems.
GeoAlign: Beyond Semantics with State-Guided Spatial Alignment in VLA Models
arXiv:2606.03240v2 Announce Type: replace Abstract: Current Vision-Language-Action (VLA) models often optimize for semantic grounding, whereas executable manipulation requires geometry-aware spatial alignment. We introduce GeoAlign, a state-guided spatial alignment architecture for VLA policy learning. Offline robot-domain RGB-D supervision post-trains an RGB geometry branch to produce Geometry-Enhanced Post-Trained (GEP) features; the depth head is then discarded, so policy training and rollout use RGB, language, and proprioceptive state. State-generated queries attend to the GEP grid to produce eight compact geometry tokens for action prediction. GeoAlign achieves 99.0% on LIBERO, 85.3% across three SimplerEnv-Fractal task families, and 78.8% on eight real-world ALOHA tasks, with ablations supporting the combined recipe of geometry post-training and state-guided querying on Isaac-GR00T N1.6-3B. Project website: https://chenyizhi123.github.io/geoalign-project/.
Abstract:Current Vision-Language-Action (VLA) models often optimize for semantic grounding, whereas executable manipulation requires geometry-aware spatial alignment. We introduce GeoAlign, a state-guided spatial alignment architecture for VLA policy learning. Offline robot-domain RGB-D supervision post-trains an RGB geometry branch to produce Geometry-Enhanced Post-Trained (GEP) features; the depth head is then discarded, so policy training and rollout use RGB, language, and proprioceptive state. State-generated queries attend to the GEP grid to produce eight compact geometry tokens for action prediction. GeoAlign achieves 99.0% on LIBERO, 85.3% across three SimplerEnv-Fractal task families, and 78.8% on eight real-world ALOHA tasks, with ablations supporting the combined recipe of geometry post-training and state-guided querying on Isaac-GR00T N1.6-3B. Project website: this https URL.
Robotics (cs.RO)
arXiv:2606.03240 [cs.RO]
(or arXiv:2606.03240v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.03240 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 · 205 words
Related coverage
GeoAlign: Beyond Semantics with State-Guided Spatial Alignment in VLA Models
arXiv · Jun 3, 2026
Beyond Isolated Entities: Relation-Aware Multi-Entity Modeling for Unsupervised Video Anomaly Detection
arXiv · Sep 29, 2026
TacWAM: Anchor-Guided World Action Model with Mechanics-Aware Tactile Prediction
arXiv · Jul 31, 2026
Article: The AI Productivity Paradox in Test Automation: Moving beyond Structural Validation to Perception and Intent
InfoQ · Jun 1, 2026