Theia: Distilling Diverse Vision Foundation Models for Robot Learning

Jinghuan Shang, Karl Schmeckpeper, Brandon B. May, Maria Vittoria Minniti, Tarik Kelestemur, David Watkins, Laura Herlant

Keywords: visual representation, robot learning, distillation, foundation model

TL;DR: We build a robot vision foundation model by distilling existing vision foundation models, which improves downstream robot learning performance.

Abstract: Vision-based robot policy learning, which maps visual inputs to actions, necessitates a holistic understanding of diverse visual tasks beyond single-task needs like classification or segmentation. Inspired by this, we introduce Theia, a vision foundation model for robot learning that distills multiple off-the-shelf vision foundation models trained on varied vision tasks. Theia's rich visual representations encode diverse visual knowledge, enhancing downstream robot learning. Extensive experiments demonstrate that Theia outperforms its teacher models and prior robot learning models using less training data and smaller model sizes. Additionally, we quantify the quality of pre-trained visual representations and hypothesize that higher entropy in feature norm distributions leads to improved robot learning performance. Code, models, and demo are available at https://theia.theaiinstitute.com.

Website: https://theia.theaiinstitute.com/

Code: https://github.com/bdaiinstitute/theia

Student Paper: yes

Submission Number: 171