Physical Intelligence: context-steered heterogeneous robot-policy training

What changed

Scientific scope

Conditioned heterogeneous robot and non-robot training episodes on detailed language, performance/strategy metadata, control modality and visual subgoals, supporting steerable composition rather than naive mixed-quality behavior averaging. Prompt conditioning, prior pi models and image-subgoal planning are inherited. Acquired delta is detailed-context integration of varied-quality/embodiment data and its ablations. Some demonstrations use language coaching or a trained high-level policy; no claim of unseen-task success from an unassisted one-line prompt. Original87-person snapshot retained, aliases not independently resolved.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Prompt conditioning, prior pi models and image-subgoal planning are inherited. Acquired delta is detailed-context integration of varied-quality/embodiment data and its ablations. Some demonstrations use language coaching or a trained high-level policy; no claim of unseen-task success from an unassisted one-line prompt. Original87-person snapshot retained, aliases not independently resolved.

Contributors

Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

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