Same intent. Different embodiment.
Coordinate frames and gripper semantics become a shared action representation that can be evaluated across embodiments.
Discuss a similar programmeDifferent arms, cameras and grippers encode actions differently.
The hard problem
A demonstration captured on one arm, camera layout or gripper did not transfer cleanly to another. Task labels also lacked the physical-state context needed for useful comparison.
Our advantage
We normalised observations and actions, aligned coordinate frames, annotated phases and contacts, recorded embodiment metadata and designed transfer-focused evaluation slices. The contribution was the representation and its qualification, rather than a change of robot brand.
Value delivered
The delivery combined synchronised trajectories, embodiment descriptors, normalised action representations, contact labels, transfer benchmarks and a governed release package.
Platform context.
Bring the next
requirement.