Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
CHORD (Contact Wrench Guidance from Human Demonstration) is an NVIDIA research framework for long-horizon dexterous manipulation of rigid and articulated objects with reinforcement learning. It represents human and robot motions by the forces and torques they can induce on the object, so that similarity is measured by the instantaneous object motion they produce rather than by contact position alone. Demonstrated contacts and contact wrenches are converted into rewards for training robot-hand policies.
The project also introduces a simulation benchmark of 4,739 bimanual dexterous manipulation tasks built from motion-capture datasets and reconstructed in-house videos. On 1,831 benchmark tasks, CHORD reports an average success rate of 82.12%, and it outperforms baselines such as DexMachina, ManipTrans, Spider and Human2Sim2Robot in the project's comparison table. With a whole-body motion inpainting module, it also learns humanoid whole-body manipulation from hand-only and third-person demonstrations (90.77% success), and learned policies transfer to real dexterous hands in open-loop and closed-loop settings.
CHORD is a research project accepted at CoRL 2026 and is presented on a project page. The page states the benchmark tasks are open-sourced, and a creator summary says code is coming soon. No code repository, model weights or paper link were provided in the material reviewed.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Framework teaching robots dexterous hand skills from human demonstrations by representing them as forces and torques on the object, then using reinforcement learning. Works across robot hands and articulated objects; code coming soon.
Closest neighbours by meaning, not just by category.
Research
Research method that retargets human hand-object demonstrations to dexterous robot hands using one multi-reference RL policy.
Model · NVIDIA
NVIDIA's real-time generative motion model for game characters and humanoid robots, with smart primitives for authoring.
Research
Hierarchical humanoid loco-manipulation framework that chains meta-skills via contact flow, accepted to CoRL 2026.
Model
Research framework that conditions one video world model on force, camera, hand pose and robot pose actions.
Hardware · Unitree
Human-sized dexterous robot hand from Unitree with 22 degrees of freedom and impact torque protection.
Research
Research framework letting a frozen VLM agent learn robot tasks in context from videos, images and feedback, no fine-tuning.
