Where are we in the search for an Artificial Visual Cortex for Embodied Intelligence?

TL;DR

This comprehensive empirical study evaluates pre-trained visual representations for Embodied AI across diverse tasks, revealing that larger datasets do not always lead to better performance and that task-specific adaptation significantly enhances results.

Contribution

The paper provides the largest systematic evaluation of visual foundation models for Embodied AI, including a new benchmark, CortexBench, and insights into dataset size, diversity, and adaptation effects.

Findings

Scaling dataset size does not universally improve performance.

Task-specific adaptation of VC-1 yields significant gains.

VC-1 outperforms prior models on average and in real-world tests.

Abstract

We present the largest and most comprehensive empirical study of pre-trained visual representations (PVRs) or visual 'foundation models' for Embodied AI. First, we curate CortexBench, consisting of 17 different tasks spanning locomotion, navigation, dexterous, and mobile manipulation. Next, we systematically evaluate existing PVRs and find that none are universally dominant. To study the effect of pre-training data size and diversity, we combine over 4,000 hours of egocentric videos from 7 different sources (over 4.3M images) and ImageNet to train different-sized vision transformers using Masked Auto-Encoding (MAE) on slices of this data. Contrary to inferences from prior work, we find that scaling dataset size and diversity does not improve performance universally (but does so on average). Our largest model, named VC-1, outperforms all prior PVRs on average but does not universally dominate either. Next, we show that task- or domain-specific adaptation of VC-1 leads to substantial gains, with VC-1 (adapted) achieving competitive or superior performance than the best known results on all of the benchmarks in CortexBench. Finally, we present real-world hardware experiments, in which VC-1 and VC-1 (adapted) outperform the strongest pre-existing PVR. Overall, this paper presents no new techniques but a rigorous systematic evaluation, a broad set of findings about PVRs (that in some cases, refute those made in narrow domains in prior work), and open-sourced code and models (that required over 10,000 GPU-hours to train) for the benefit of the research community.