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

TL;DR

This paper introduces MCF-Proto, a lightweight, motion-centric action head for vision-language-action models that improves robustness and organization of robotic manipulation actions through geometric and compositional structure.

Contribution

It proposes a novel action head that predicts local frames and prototypes, leading to more stable, compact, and generalizable action representations without auxiliary supervision.

Findings

Learned local frames develop stable geometric structure.

Actions become more compact and organized by shared prototypes.

Enhanced robustness under geometric perturbations.

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.