WUJI showcases Hand 2's pen spinning capabilities at IROS 2026

WUJI: WUJI, also known as Wuji Technology, develops high-dexterity robotic hands featuring multiple independent motors per finger. The company focuses on reinforcement learning techniques to train manipulation skills in physics simulators before transferring them to physical hardware. In this news, WUJI demonstrated its Hand 2 model successfully performing complex pen-spinning tasks through simulation-to-real deployment and released the supporting mjlab toolkit under an open-source license. Han Yang: Han Yang is part of the Wuji Technology team focused on robotic hand control and learning algorithms. Yang helped create the resources for training and deploying policies on the 20-joint Hand 2, including the pen-spinning application. Yang's contributions appear in the recent open-source project documentation and IROS 2026 showcase. Guanqi He: Guanqi He is a member of the Wuji Technology team developing high-performance robotic hands and training software. He helped produce the assets and guides for the Hand 2 pen-spinning experiment released under Apache 2.0. He's work supports the project's emphasis on accessible simulation-to-real robotics research. Jielin Wu: Jielin Wu is a researcher affiliated with Wuji Technology contributing to projects on robotic dexterous manipulation. Wu co-authored the Wuji-MJLab work that enables reinforcement learning training for in-hand tasks such as pen spinning. Wu's involvement centers on the simulation-based policy development and real-world validation shown in the recent demonstration. Lijie Sheng: Lijie Sheng works with Wuji Technology on projects advancing robotic dexterity through learning-based methods. Sheng contributed to the Wuji-MJLab release that includes training environments and deployment code for Hand 2 tasks. Sheng's involvement aligns with the team's focus on practical, reproducible robot manipulation research. Shenzhe Yao: Shenzhe Yao is a researcher at Wuji Technology who participated in the development of the mjlab robot-learning toolkit. Yao contributed to the open-source release of models, motion data, and deployment guides for the WUJI Hand. Yao's work supports the team's efforts in advancing simulation-trained policies for physical robot hands. Xiaohan Liu: Xiaohan Liu is a researcher associated with Wuji Technology and the mjlab toolkit. Liu participated in creating the reinforcement learning setups and sensor integration shown in the pen-spinning and glove-control demonstrations. Liu's contributions are part of the recent open-source publication and conference presentation. Li Chengmeng: Li Chengmeng is a researcher at Wuji Technology who aided in the development and release of tools for robotic hand control. Li helped prepare the motion clips, models, and hardware guides included in the open-source bundle. Li's efforts support broader adoption of the simulation-to-real techniques demonstrated at IROS 2026. Wentao Zhang: Wentao Zhang contributes to Wuji Technology's research on dexterous robotics and simulation training pipelines. Zhang participated in the Wuji-MJLab project that produced the trained policies transferred from virtual environments to the physical hand. Zhang's work features in the documentation for replicating the pen-spinning setup. Xiangrui Jiang: Xiangrui Jiang is a researcher with Wuji Technology involved in the mjlab framework and hand manipulation experiments. Jiang supported the integration of motion tracking and reinforcement learning methods used in the pen-spinning demo. Jiang's role ties directly to the open-source assets released alongside the Hand 2 results. Open Source Release: WUJI made the full training code, motion data, models, and printable files available under the Apache 2.0 license for researchers and developers. Conference Presentation: The robotic hand demonstration and supporting tools were showcased at IROS 2026, highlighting advances in simulation-trained dexterous manipulation. Human-in-the-Loop Teaching: A sensor-equipped glove with fingertip tracking and palm pressure sensing provides an alternative method for collecting human demonstration data to train robot hands.