WRSWorld Research StateContribute
WORLD RESEARCH STATE

Physical Intelligence: heterogeneous semantic/action co-training for robot transfer

Contribution grade

Vision-language-action transfer systems · Human-designed research method or resource

Reference score125,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Co-trained on robot trajectories, web vision-language tasks, object grounding and semantic subtask prediction, combining high-level instructions and a low-level action expert to transfer manipulation to unseen household environments.

Scientific scope

Co-trained on robot trajectories, web vision-language tasks, object grounding and semantic subtask prediction, combining high-level instructions and a low-level action expert to transfer manipulation to unseen household environments. Uses prior pi0 architecture. Acquired delta is heterogeneous-task/hybrid-example integration and transfer evidence, not another size or invention of VLA. Evaluated new homes are a finite test set, not all open worlds; later knowledge-insulation method has separate original credit.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Uses prior pi0 architecture. Acquired delta is heterogeneous-task/hybrid-example integration and transfer evidence, not another size or invention of VLA. Evaluated new homes are a finite test set, not all open worlds; later knowledge-insulation method has separate original credit.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction