Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models

2026-05-12

AI Summary

Existing vision-language-action (VLA) models typically predict actions directly in a fixed world coordinate frame, lacking explicit modeling of motion geometry, which limits policy generalization and robustness. This work proposes MCF-Proto, a lightweight action head that introduces a motion-centric frame (MCF) and a prototype-based action parameterization. Without requiring additional supervision, it adaptively learns a local coordinate frame aligned with end-effector motion and achieves a compact, structured action representation through shared prototypes. By integrating SO(3) rotation transformations, prototype clustering, and local-to-world coordinate mapping, the method enables end-to-end training and significantly enhances policy robustness under geometric perturbations and behavioral generalization, yielding fewer and more regularly organized dominant action directions.

Abstract

Vision-Language-Action (VLA) models have advanced rapidly with stronger backbones, broader pre-training, and larger demonstration datasets, yet their action heads remain largely homogeneous: most directly predict action commands in a fixed world coordinate frame. We propose \textbf{MCF-Proto}, a lightweight action head that equips VLA policies with a Motion-Centric Action Frame (MCF) and a prototype-based action parameterization. At each step, the policy predicts a rotation $R_t \in SO(3)$, composes actions in the transformed local frame from a set of prototypes, and maps them back to the world frame for end-to-end training, using only standard demonstrations without auxiliary supervision. This simple design induces stable emergent structure. Without explicit directional labels, the learned local frames develop a stable geometric structure whose axes are strongly compatible with demonstrated end-effector motion. Meanwhile, actions in the learned representation become substantially more compact, with variation captured by fewer dominant directions and more regularly organized by shared prototypes. These structural properties translate into improved robustness, especially under geometric perturbations. Our results suggest that adding lightweight geometric and compositional structure to the action head can materially improve how VLA policies organize and generalize robotic manipulation behavior. An anonymized code repository is provided in the supplementary material.

Problem

• Vision-Language-Action models
• action head
• coordinate frame
• geometric structure
• robustness

Innovation

• Motion-Centric Action Frame
• Prototype-based Action Parameterization
• Vision-Language-Action Models
• Geometric Structure Emergence
• Robotic Manipulation Generalization