The Toughest Dexterous Robotic Hand Yet: OmniHand 3 Ultra‑T, Lite, and OmniPicker 3 Unveiled

Recent humanoid robot demos showcase impressive locomotion, yet high‑intensity actions often omit the hands because traditional dexterous hands are fragile, expensive, and hard to maintain. The article explains that a robust hand is essential for tasks such as grasping, assembly, and manipulation, and that the industry has been split between two main actuation strategies: direct drive and rope drive.

Technical Routes

Direct drive attaches motors to each joint, offering fast response, high control precision, and direct force feedback, but adds weight, volume, and heat, making high‑DOF hands bulky. Rope drive moves the motors away from the hand, using tendon‑like structures to pull fingers, resulting in lighter, more flexible hands better suited for complex, unstructured environments. However, rope drive introduces control challenges due to elasticity, backlash, and wear.

Rather than choosing one path, the company pursues both in parallel, integrating rope‑drive as the primary architecture while adding selective direct‑drive capability for critical joints, creating a hybrid system.

OmniHand 3 Ultra‑T (Flagship)

Targeted at both research and industrial deployment, the Ultra‑T is positioned as the first industrial‑grade rope‑driven dexterous hand. It boasts five "super" attributes: ultra‑flexibility, ultra‑human likeness, ultra‑strength, ultra‑reliability, and ultra‑perception.

Flexibility: 22+3 DOF bionic configuration with a thumb offering four active DOF and a palm with one active DOF; achieves a perfect score on the Kapandji test.

Speed: 0.3 s open‑close cycle, enabling both fine manipulation and line‑speed operation.

Weight: Hand mass under 500 g (magnesium alloy frame, titanium‑reinforced stress points), the lightest 22+3 DOF solution globally.

Range of Motion: Wrist curvature 55° front‑back and 40° left‑right, surpassing the industry typical <30°.

Strength: In‑house planetary‑roller‑screw micro‑actuator delivering 300 N rated output, 10:1 load‑to‑self‑weight ratio; tendon break strength 1000 N.

Reliability: Quick‑release tendon mechanism reduces replacement time to <10 min, supporting 7×24 h continuous operation.

Perception: Integrated vision‑touch sensor suite with an internal palm camera provides full‑hand 3‑D tactile perception, eliminating visual occlusion during grasp.

The hand combines rope‑drive with selective direct‑drive joints, improving finger side‑swing torque, control accuracy, and long‑term maintainability while preserving lightweight advantages.

OmniHand 3 Lite (Affordable)

Designed for cost‑sensitive applications, the Lite version is the world’s smallest dexterous hand (mouse‑sized) yet maintains industrial‑grade durability. It features an aerospace‑grade metal skeleton with integrated silicone, passes rigorous impact, drop, and collision tests, and supports modular micro‑joint modules, optional fisheye camera, and full‑hand tactile coverage. It is positioned as the standard hand for 0.8–1.3 m semi‑size humanoid robots.

Price is kept below US$300, enabling academic labs and small teams to build complete manipulation and perception setups without prohibitive expense.

OmniPicker 3 (Industrial Gripper)

The upgraded gripper emphasizes industrial reliability: 1 M cycles with load, 140 N gripping force, 0.4 s open‑close speed, and 0.1 N tactile force resolution. It supports modular finger adapters for diverse object shapes and integrates a fully compatible UMI data acquisition device, allowing direct transfer of force and pose data for reinforcement‑learning pipelines.

Production Scale and Ecosystem

Since its founding in January 2026, the company has delivered over 8 000 dexterous hands and 10 000+ grippers, with ~1 000 units deployed in data‑collection factories across retail, pharmacy, and manufacturing sites, achieving stable 8‑hour continuous operation and accumulating >50 000 h of real‑world data.

All products share a unified open‑source hardware‑software stack, supporting RL training, ROS/ROS2, NVIDIA Isaac Gym, and MuJoCo, lowering the barrier for developers and researchers to focus on algorithms rather than low‑level integration.

Conclusion

The launch demonstrates that the industry’s toughest challenge—building durable, high‑performance dexterous hands—can be addressed without sacrificing speed, strength, or affordability. By delivering a reusable technology platform rather than a single product, the company aims to accelerate embodied intelligence research and industrial adoption.