SOMA Retargeter transforms human demonstrations into viable joint motions by adjusting scale, solving poses, and correcting for mechanical constraints. The pipeline accepts Biovision Hierarchy animations standardized on the SOMA base skeleton, drawing from motion capture datasets, video reconstructions, or generative models. It outputs detailed CSV trajectory files featuring root positions, orientations, and frame-by-frame joint angles to steer robotic control systems. A built-in viewer also allows developers to visually inspect actions like forward walking and compare retargeted hardware motions against original human data.
Key points
The v0.2 release supplies reference configurations and retargeting results tailored for f…: The v0.2 release supplies reference configurations and retargeting results tailored for five humanoid platforms: Unitree G1, Unitree H2, Booster T1, AgiBot X2 Ultra, and AgiBot A3T3.
An integrated viewer enables engineers to visually verify retargeted movements against or…: An integrated viewer enables engineers to visually verify retargeted movements against original human demonstration animations.
The software eliminates the impracticality of direct human motion copying by programmatic…: The software eliminates the impracticality of direct human motion copying by programmatically accounting for differing joint architectures and ranges of motion.
Sources checked by our analysis pipeline: 1
Human motion offers roboticists a great starting point for teaching humanoid robots how to move. A human demonstration captures the coordination and timing behind natural movement, from everyday actions to complex tasks. But a robot cannot simply copy the same joint motion from a human. Its proportions, joint structure, and range of motion are different, so the movement needs to be adapted to fit the robot.
The open source NVIDIA SOMA Retargeter helps you adapt human motion to your robot’s proportions and range of motion. Built around the shared human-motion representation in NVIDIA SOMA-X, it provides tools to configure your robot, refine the resulting movement, and process motion data for downstream training.
This post explains the retargeting process, the tools for refining robot motion, and the connection to policy learning.
From human motion to robot joint trajectories
Human motion can come from many sources, including a motion capture session, a video, an online dataset, or a generative model. These sources often output motion with different skeletons and file formats. Representing compatible human motion on the SOMA skeleton gives the retargeter…
Designed to optimize ergonomic comfort across varied hand sizes, the newly introduced peripheral lets users mechanically tailor the thumb interface rather than forcing them to adapt to a fixed shape. This mechanical customization is aimed at reducing physical strain during prolonged computer use. Available in three color choices, the hardware offers a budget-friendly option for desktop users seeking flexible input controls.
C++ engineers seeking cleaner coroutine syntax can avoid the friction of manually defining combined parameter types by utilizing dedicated helper functions. Introducing a streamlined method such as `start()` bundles critical inputs, including execution priority, directly into an awaitable operation. This design pattern reduces boilerplate overhead and ensures that standard asynchronous invocation remains intuitive for consumers of the class.
Meta’s Reality Labs engineers completed the transition to unify their development environments while building non-phone devices like AI glasses. The team previously faced doubled tooling costs and hidden bugs caused by splitting testing between macOS emulators and Linux platforms. By moving the Android Open Source Project’s high-fidelity framework directly to macOS, engineers can now diagnose platform-level issues on local hardware.