Capturing Visual Environment Structure Correlates with Control Performance

TL;DR

This paper demonstrates that probing pretrained visual encoders for environment state decoding correlates strongly with control policy performance, offering a new efficient metric for selecting visual representations in robotic manipulation tasks.

Contribution

It introduces a novel probing method to evaluate visual encoders based on their support for decoding environment state, improving representation selection for control.

Findings

Probing accuracy correlates with downstream policy success.

The method outperforms existing proxy metrics.

Encoding physical environment state enhances control generalization.

Abstract

The choice of visual representation is key to scaling generalist robot policies. However, direct evaluation via policy rollouts is expensive, even in simulation. Existing proxy metrics focus on the representation's capacity to capture narrow aspects of the visual world, like object shape, limiting generalization across environments. In this paper, we take an analytical perspective: we probe pretrained visual encoders by measuring how well they support decoding of environment state -- including geometry, object structure, and physical attributes -- from images. Leveraging simulation environments with access to ground-truth state, we show that this probing accuracy strongly correlates with downstream policy performance across diverse environments and learning settings, significantly outperforming prior metrics and enabling efficient representation selection. More broadly, our study provides insight into the representational properties that support generalizable manipulation, suggesting that learning to encode the latent physical state of the environment is a promising objective for control.