LARY: A Latent Action Representation Yielding Benchmark for Generalizable Vision-to-Action Alignment

TL;DR

LARY introduces a comprehensive benchmark to evaluate latent action representations derived from large-scale human videos, revealing that general visual models outperform specialized ones in vision-to-action tasks.

Contribution

The paper presents the LARY benchmark dataset and framework, enabling rigorous evaluation of latent action representations across semantic and control tasks, highlighting the superiority of general visual models.

Findings

General visual foundation models outperform specialized embodied models.

Latent-based visual space aligns better with physical action space than pixel-based space.

Semantic abstraction from vision is more effective for control than pixel-level reconstruction.

Abstract

While the shortage of explicit action data limits Vision-Language-Action (VLA) models, human action videos offer a scalable yet unlabeled data source. A critical challenge in utilizing large-scale human video datasets lies in transforming visual signals into ontology-independent representations, known as latent actions. However, the capacity of latent action representation to derive robust control from visual observations has yet to be rigorously evaluated. We introduce the Latent Action Representation Yielding (LARY) Benchmark, a unified framework for evaluating latent action representations on both high-level semantic actions (what to do) and low-level robotic control (how to do). The comprehensively curated dataset encompasses over one million videos (1,000 hours) spanning 151 action categories, alongside 620K image pairs and 595K motion trajectories across diverse embodiments and environments. Our experiments reveal two crucial insights: (i) General visual foundation models, trained without any action supervision, consistently outperform specialized embodied latent action models. (ii) Latent-based visual space is fundamentally better aligned to physical action space than pixel-based space. These results suggest that general visual representations inherently encode action-relevant knowledge for physical control, and that semantic-level abstraction serves as a fundamentally more effective pathway from vision to action than pixel-level reconstruction.