PhysBrain 1.0 Technical Report
For embodied AI researchers, this work provides a method to bridge multimodal understanding and robot action by scaling physical commonsense from human video, though it is incremental as it combines existing techniques.
PhysBrain 1.0 converts large-scale human egocentric video into structured physical commonsense supervision via a data engine, then transfers these priors to VLA policies, achieving SOTA results on ERQA, PhysBench, SimplerEnv-WidowX, LIBERO, and RoboCasa, with strong out-of-domain performance on SimplerEnv.
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.