Tars DexHand Debuts at ICRA 2026, Showcasing 'Mind to Hand' Integration

ME AI News: On June 1, local time in Austria, the International Conference on Robotics and Automation (ICRA) officially opened in Vienna. As one of the most influential international conferences in the field of robotics and automation, ICRA annually brings together top talent from academia, industry, and cutting-edge technology companies. At this year’s conference, itshizhihang presented under the theme “Mind to Hand,” showcasing a complete closed-loop system spanning data acquisition, perception and decision-making, to hardware embodiment—becoming a highly visible representative of Chinese embodied intelligence companies at the exhibition. The event marked the official debut of itshizhihang’s TARS DexHand dexterous hand. Its core breakthrough lies not merely in isolated hardware performance, but in achieving high-level integration of hardware, data, and model capabilities to form an entirely new “hand-brain unified” system. On the hardware side, the DexHand dexterous hand employs a 21-degree-of-freedom quasi-direct-drive design that precisely replicates human hand bone structure and joint distribution, enabling it to match human-level capabilities in joint coordination and high-precision motion combinations, while also ensuring smooth motion, high accuracy, and production consistency. On the data side, the DexHand dexterous hand deeply aligns with human-centric data paradigms, enabling high-fidelity mapping of human motion data onto the hand—significantly improving the utilization efficiency of embodied data. On the model side, the DexHand dexterous hand integrates with itshizhihang’s general embodied large model AWE 3.0 and leverages TacForeSight, a newly launched technology led by itshizhihang, to direct wrist force/torque signals to the visuo-tactile world model OmniVTA. As a result, the dexterous hand can not only perceive physical textures such as roughness, softness, and hardness, but also proactively predict physical world dynamics and adjust its action strategies—compressing perception, understanding, prediction, and manipulation into a single closed loop to truly realize “hand-brain unity.” At the exhibition, the DexHand dexterous hand demonstrated its operational capability through sign language interactions involving all 26 English letters. Visitors could type words like “TARS” or “ROBOT” on a screen, and the hand would fluidly perform the corresponding gestures—drawing significant crowds. This not only showcased the precision of DexHand’s hand posture control but also vividly illustrated its hand-brain collaborative ability—from command comprehension to action execution—aligning with ICRA 2026’s theme “Robots for All” and embodying the inclusive vision of embodied intelligence connecting diverse populations. The general embodied large model AWE also made its first overseas offline debut. Powered by AWE 3.0, itshizhihang’s A1 robot demonstrated real-life fine-grained packing tasks throughout the day, successfully completing over seven consecutive long-range steps—including picking up a pencil, placing it in a pencil case, zipping a bag, and organizing books—each step dynamically determined based on the current object pose and contact state, rather than following pre-programmed trajectories. Even more technically challenging was the sub-millimeter cable connector task: after randomly adjusting the cable port positions on-site, the A1 robot autonomously corrected its path and successfully inserted the connector using AWE 3.0’s real-time perception and re-planning capabilities—all without any human intervention. One scholar who observed the demonstration remarked that this self-correction ability represents precisely the kind of human-like intelligence that is most valuable—and rare—in industrial settings. As a premier academic conference in robotics and automation, ICRA is not merely a showcase for cutting-edge products but also a critical platform for evaluating technological and research excellence. Following its victory in the ICRA 2025 ManiSkill-ViTac vision-tactile fusion challenge by defeating 42 global teams, nine research papers led by Dr. Ding Wenchao and his collaborators from itshizhihang and partner institutions were accepted into ICRA 2026—the highest number among all invited speakers at the official forum. These studies span robot vision perception and understanding, viewpoint-robust manipulation, and complex task generalization. This rich and rigorous academic output is closely tied to itshizhihang’s high talent density, long-term academic accumulation, and engineering exploration. To date, over 80% of its workforce consists of R&D personnel with master’s or doctoral degrees; the team has collectively published more than 40 papers in robotics-related fields, continuously driving the translation of frontier research into practical capabilities. (Source: Ifnar)