SOMA Retargeter adapts human movement without claiming robot balance
NVIDIA SOMA Retargeter adapts human motion to humanoid robots with different proportions and joint limits. It scales targets, solves inverse kinematics and applies corrections such as reducing foot sliding, then exports joint trajectories. These are movement references for later training or control; an animation that resembles a human walk does not establish that a physical robot can execute it stably.
NVIDIA Technical Blog · Nov 13, 2026
Why it matters
Motion conversion and successful robot control need separate validation rather than a single convincing animation.
More context
The open-source v0.2 examples include five humanoid configurations. Install from the repository and try the bundled Unitree G1 walk in the viewer before processing larger collections. Downstream simulation should check balance, contact, actuator limits and tracking accuracy.
Keep in mind
Available developer tooling with experimental weight optimization. Input motion must use the supported SOMA representation, and retargeted trajectories are not deployment-ready controllers or evidence of safe physical execution.
Original read