Rethinking Visual-Language-Action Model Scaling: Alignment, Mixture, and Regularization

TL;DR

This paper systematically investigates how scaling Vision-Language-Action models affects robotic control, revealing critical factors like physical alignment and data heterogeneity, and providing practical training insights.

Contribution

It offers a comprehensive analysis of VLA model scaling in robotics, highlighting the importance of physical alignment, the risks of data pooling, and the limited impact of regularization strategies.

Findings

Unified end-effector relative action representation improves cross-embodiment transfer.

Naive pooling of heterogeneous robot data can cause negative transfer.

Regularization strategies like sensory dropout do not consistently enhance performance.

Abstract

While Vision-Language-Action (VLA) models show strong promise for generalist robot control, it remains unclear whether -- and under what conditions -- the standard "scale data" recipe translates to robotics, where training data is inherently heterogeneous across embodiments, sensors, and action spaces. We present a systematic, controlled study of VLA scaling that revisits core training choices for pretraining across diverse robots. Using a representative VLA framework that combines a vision-language backbone with flow-matching, we ablate key design decisions under matched conditions and evaluate in extensive simulation and real-robot experiments. To improve the reliability of real-world results, we introduce a Grouped Blind Ensemble protocol that blinds operators to model identity and separates policy execution from outcome judgment, reducing experimenter bias. Our analysis targets three dimensions of VLA scaling. (1) Physical alignment: we show that a unified end-effector (EEF)-relative action representation is critical for robust cross-embodiment transfer. (2) Embodiment mixture: we find that naively pooling heterogeneous robot datasets often induces negative transfer rather than gains, underscoring the fragility of indiscriminate data scaling. (3) Training regularization: we observe that intuitive strategies, such as sensory dropout and multi-stage fine-tuning, do not consistently improve performance at scale. Together, this study challenge some common assumptions about embodied scaling and provide practical guidance for training large-scale VLA policies from diverse robotic data. Project website: https://research.beingbeyond.com/rethink_vla