WUJI HAND 2
About
About WUJI HAND 2
WUJI HAND 2 is a dexterous robotic hand built around a biomimetic skeleton and a soft outer body. Four active joints per finger provide thumb opposition and lateral finger movement. Back-drivable rotary actuation, hybrid force-position control and multi-axis fingertip sensing let developers work with contact forces as well as finger poses. Modular construction supports replacement of individual assemblies.
The hand connects to robot hosts over Ethernet. Wuji’s Python and C interfaces, ROS 2 support, simulation models and glove retargeting tools cover device control, teleoperation and manipulation-policy development. The Wuji-MJLab package runs learned cube-reorientation and reference-tracking pen-spinning policies on physical Hand 2 hardware. External experiments also use the hand on dual-arm research systems and mobile manipulators.
Capabilities
Source-backed capabilities
- Operating environments: Physical research setup
- Control modes: Autonomous Diffusion Policy, Learned Policy
- Drawer opening and cube placement
Research demonstration - 2026-09-30 - Hook a drawer open, place a cube with a thumb–index pinch, then close the drawer with a fist. - environment: Physical research setup
Source - Toy pick-and-place
Research demonstration - 2026-09-30 - Pick up a toy and place it in a basket with a policy learned from DITTO-X demonstrations. - environment: Physical research setup
Source - Pick and place
Research demonstration - 2026-09-30 - Wuji Hand 2 performs pick-and-place color sorting from a moving conveyor using the right Flexiv Rizon 4S arm; the static left arm supplies another camera view.
Source - Cube reorientation
Research demonstration - 2026-09-27 - Rotate a tagged cube within the hand using a learned policy and camera feedback. - environment: Physical research setup
Source - Pen spinning
Research demonstration - 2026-09-27 - Spin and reorient a tagged pen with a reference-tracking policy trained in simulation. - environment: Physical research setup
Source - Dexterous manipulation
Research demonstration - 2026-08-31 - Wuji Hand 2 tears paper in dexterous manipulation tests using the Tianji Gento Luna research setup.
Source - Remote and shared control
Research demonstration - 2026-09-30 - Researchers teleoperate Hand 2 fingers with a Manus glove and map Vive-tracked wrist motion to the arm during real demonstration collection.
Source
Observed use
Exact-product configurations are shown without inferring supply, purchase, customer status, standard fit, deployment or partnership.
- Agile ONE SN
2026-10-21 - Reference Design - Mounted - Physical - Agile ONE SN was announced with two WUJI Hand 2 dexterous hands in its research reference configuration, with platform availability planned for early 2027.
Source - DITTO-X experimental arm setup
Research Experiment - Mounted - Physical - DITTO-X uses Wuji Hand 2 for drawer-and-cube manipulation and toy pick-and-place in a physical dexterous teleoperation study.
Source - Wuji-MJLab training and deployment stack
Research Experiment - Controlled By - Physical - Wuji’s Wuji-MJLab stack runs simulation-trained policies on physical Hand 2 for cube reorientation and reference-tracking pen spinning, with dedicated hand jigs, vision and deployment tools.
Source - Tianji Gento Luna dual-arm research setup
Research Experiment - Mounted - Physical - Motus2 researchers mounted Wuji Hand 2 on Tianji Gento Luna for dual-arm data collection and post-training, including a physical tactile-feedback paper-tearing experiment.
Source - Rizon 4S
Research Experiment - Mounted - Physical - Industrial Next used Wuji Hand 2 on Flexiv Rizon 4S arms for conveyor pick-and-sort research. The right arm picked and sorted objects; the left remained static with an additional camera.
Source - TRON 2
Physical Demo - Mounted - Physical - LimX demonstrated TRON 2 with Wuji Hand 2 in a traditional-medicine pharmacy workflow spanning picking, weighing, grinding and packaging across cabinets and counters.
Source
Supply
Supply / Components
- WUJI HAND 2
Dexterous robotic hands
Research
- OATH: Object-Agnostic Tracking for Dexterous Hands, Enabling Rapid Physics-based Retargeting
2026-10-09 - Research Publication - OATH uses a contact-conditioned residual policy to retarget human grasping trajectories to dexterous hands across unseen objects. Multiply Labs and university researchers report 58% held-out and 54% unseen-object trajectory recovery in DexYCB simulation, plus open-loop playback on a UR5e with Wuji Hand 2.
Source - Forward and Reverse Teleoperation for Dexterous Manipulation and Human Intervention
2026-09-30 - Research Publication - An actuated exoskeleton renders force and vibrotactile feedback during forward teleoperation. Reverse teleoperation aligns the operator’s hand with the policy-controlled robot before takeover. The described DITTO-Human collection spans six tasks, three hands and more than 16 hours; public downloads remain pending.
Source - Getting Out and Getting Back: World and Behavior Grounding in Real2Sim2Real Co-Training
2026-09-30 - Research Publication - This Industrial Next and Cambridge study separates physical-world fidelity from human-motion fidelity in simulation co-training. A dual Flexiv Rizon 4S setup with Wuji Hand 2 performs conveyor pick-and-sort. For the 500M flow-matching policy, fully grounded co-training raises success from 26/50 to 43/50 trials; the study also evaluates post-trained foundation models.
Source - Wuji-MJLab: Cube Reorientation and Pen Spinning
2026-09-27 - Research Publication - Cube reorientation supports both Wuji hand generations with separate hardware and checkpoint configurations. Reference-tracking pen spinning targets Wuji Hand 2 and transfers learned policies to the physical hand. The asset bundle includes deployment checkpoints, hand jigs, cube and pen geometry, and 40 motion clips.
Source - Motus2: A Self-Evolving General World Model for Dexterous Manipulation
2026-08-31 - Research Publication - Motus2 combines policy generation, action-conditioned world simulation and outcome evaluation in one shared model for dexterous manipulation. GensPI and Tsinghua researchers evaluate multiple dual-arm configurations, including Wuji Hand 2 on Tianji Gento Luna. In its tactile ablation, paper tearing with Hand 2 improves from 55% to 70% over 20 rollouts.
Source
Timeline
- Agile Robots unveils Agile ONE SN humanoid reference platform
2026-10-21 - Product Launch - The 1.4 m platform has 31 body degrees of freedom plus 20 in each hand and supports up to 3 kg payload per arm. GR00T 1.7, whole-body control and URDF/USD assets connect demonstration capture, simulation, training and physical testing. Availability is scheduled for early 2027; NVIDIA Halos safety integration is being explored.
Source - Industrial Next study evaluates WUJI HAND 2 conveyor pick-and-sort
2026-09-30 - Research Publication - Wuji Hand 2 performs pick-and-place color sorting from a moving conveyor using the right Flexiv Rizon 4S arm; the static left arm supplies another camera view. The 500M flow-matching policy succeeds in 26 of 50 real-only trials and 43 of 50 trials after co-training with 100 real and approximately 1,500 grounded simulated trajectories. Researchers teleoperate Hand 2 fingers with a Manus glove and map Vive-tracked wrist motion to the arm during real demonstration collection.
Source - WUJI HAND 2 gains open-source cube-reorientation and pen-spinning tools
2026-09-27 - Research Publication - Hand 2 cube reorientation includes standard and 50 Hz tasks, with separate hardware mounting and deployment configurations. Reference-tracking pen spinning includes 40 motion clips, five recipes and five matched pretrained checkpoint sets. The release bundles ONNX policies, hand mounting jigs, cube and pen CAD files, assembly drawings and tag assets.
Source - Motus2 study evaluates tactile paper tearing with WUJI HAND 2
2026-08-31 - Research Publication - Wuji Hand 2 tears paper in dexterous manipulation tests using the Tianji Gento Luna research setup. Paper-tearing success rises from 55% without tactile feedback to 70% with it, over 20 rollouts per condition. The model uses a tactile expert for contact-aware action refinement and auxiliary future-force prediction.
Source
Context
Comparable Systems
- Aero Hand Open
Chestnut Robotics - Dexterous Hand
- OmniHand O10 2025
AgiBot - Hand
- OmniHand Pro 2025
AgiBot - Hand
- AIDIN-Hand
Aidin Robotics - Robotic Hand
- Amazing Hand
Pollen Robotics - Robotic Hand
- Apex Hand
Dexcel Robotics - Dexterous Hand