PhysBrain 1.0 Technical Report

TL;DR

PhysBrain 1.0 leverages large-scale egocentric videos to learn physical commonsense, enhancing robot understanding and control across diverse benchmarks with state-of-the-art results.

Contribution

It introduces a novel data engine that converts egocentric videos into structured supervision for training physical priors in vision-language models.

Findings

Achieves SOTA on multiple benchmarks including ERQA, PhysBench, and RoboCasa.

Demonstrates strong out-of-domain performance, especially on SimplerEnv.

Shows that scaling physical commonsense from videos improves robot action understanding.

Abstract

Vision-language-action models have advanced rapidly, but robot trajectories alone provide limited coverage for learning broad physical understanding. PhysBrain 1.0 studies a complementary route: converting large-scale human egocentric video into structured physical commonsense supervision before robot adaptation. Our data engine extracts scene elements, spatial dynamics, action execution, and depth-aware relations, then turns them into question-answer supervision for training PhysBrain VLMs. The resulting physical priors are further transferred to VLA policies through a capability-preserving and language-sensitive adaptation design. Across multimodal QA benchmarks and embodied control benchmarks, including ERQA, PhysBench, SimplerEnv-WidowX, LIBERO, and RoboCasa, PhysBrain 1.0 achieves SOTA results and shows especially strong out-of-domain performance on SimplerEnv. These results suggest that scaling physical commonsense from human interaction video can provide an effective bridge from multimodal understanding to robot action.