Human demonstrations for robot training.

Convert captured, extracted, or generated human motion into joint representations built for a target robot embodiment.

Human motion does not transfer directly

A human demonstration contains useful behavior, but a robot cannot copy it joint for joint. Human and robot bodies differ in proportions, joint axes, degrees of freedom, range limits, and dynamics. A valid human pose may be unreachable or unstable for the target machine.

Manual retargeting can adapt an individual sequence. It does not scale cleanly across large demonstration sets, changing robot designs, or multiple embodiments. Every new mapping adds another body-specific preprocessing step.

One representation, multiple bodies

Uthana converts source motion into a shared 3D representation. The source can come from marker-based capture, motion extracted from video, or motion generated from text. The same input layer supports different motion sources without rebuilding the pipeline around each one.

Uthana then maps the motion to the target robot’s joint structure. The target specification defines the skeleton, joint convention, limits, timing, and any additional constraints for the engagement.

Build demonstration sets across embodiments

A shared source representation makes it possible to process a motion collection against more than one target body. Teams can reuse the underlying human behavior while generating a distinct joint sequence for each embodiment.

Scale and batch-processing workflows are scoped per engagement, against the intended robot and output specification.

Define the delivered package

Each engagement begins with the robot description, target environment, accepted joint structure, and downstream training requirements. The delivered package is defined for the engagement.

Robot description formats, output structures, and validation methods are defined for the engagement. A representative sample can be tested inside the customer’s training stack before a larger delivery is scoped.