aiqbench analysis
This paper proposes AnchorRefine, a hierarchical framework for Vision-Language-Action models that separates global trajectory planning from local execution refinement to improve robotic manipulation precision. By factorizing action modeling into an anchor planner and a residual correction module, the approach addresses limitations in monolithic VLA policies where large motions often suppress critical small-scale adjustments. Although the authors have withdrawn this specific manuscript pending further validation, the methodology represents a significant conceptual advance for improving geometric accuracy in embodied AI systems.
AnchorRefine: Synergy-Manipulation Based on Trajectory Anchor and Residual Refinement for Vision-Language-Action Models
arXiv:2604.17787v2 Announce Type: replace-cross Abstract: Precision-critical manipulation requires both global trajectory organization and local execution correction, yet most vision-language-action (VLA) policies generate actions within a single unified space. This monolithic formulation forces macro-level transport and micro-level refinement to be optimized under the same objective, causing large motions to dominate learning while suppressing small but failure-critical corrective signals. In contrast, human manipulation is structured by global movement planning together with continuous local adjustment during execution. Motivated by this principle, we propose AnchorRefine, a hierarchical framework that factorizes VLA action modeling into trajectory anchor and residual refinement. The anchor planner predicts a coarse motion scaffold, while the refinement module corrects execution-level deviations to improve geometric and contact precision. We further introduce a decision-aware gripper refinement mechanism to better capture the discrete and boundary-sensitive nature of gripper control. Experiments on LIBERO, CALVIN, and real-robot tasks demonstrate that AnchorRefine consistently improves both regression-based and diffusion-based VLA backbones, yielding gains of up to 7.8% in simulation success rate and 18% in real-world success rate.
Abstract:Precision-critical manipulation requires both global trajectory organization and local execution correction, yet most vision-language-action (VLA) policies generate actions within a single unified space. This monolithic formulation forces macro-level transport and micro-level refinement to be optimized under the same objective, causing large motions to dominate learning while suppressing small but failure-critical corrective signals. In contrast, human manipulation is structured by global movement planning together with continuous local adjustment during execution. Motivated by this principle, we propose AnchorRefine, a hierarchical framework that factorizes VLA action modeling into trajectory anchor and residual refinement. The anchor planner predicts a coarse motion scaffold, while the refinement module corrects execution-level deviations to improve geometric and contact precision. We further introduce a decision-aware gripper refinement mechanism to better capture the discrete and boundary-sensitive nature of gripper control. Experiments on LIBERO, CALVIN, and real-robot tasks demonstrate that AnchorRefine consistently improves both regression-based and diffusion-based VLA backbones, yielding gains of up to 7.8% in simulation success rate and 18% in real-world success rate.
Robotics (cs.RO); Artificial Intelligence (cs.AI)
arXiv:2604.17787 [cs.RO]
(or arXiv:2604.17787v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2604.17787 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 · 258 words
Related coverage
VOMMI: Collecting and Leveraging Portable Demonstrations for Mobile Manipulation
arXiv · Oct 7, 2026
MomADv2: Reliable Temporal Memory for End-to-End Autonomous Driving
arXiv · Aug 25, 2026
Residual Compliant Policy: Learning Compliant Policies from Few Demonstrations
arXiv · Oct 14, 2026
MobiGuide: Vocabulary-Guided Online Recovery for Mobile Manipulation
arXiv · Oct 15, 2026