Gasgoo Munich- ROBOTERA released its XHAND 1 Pro-U, a high-degree-of-freedom five-finger dexterous hand, on November 16. Available in limited quantities to global research teams and embodied AI developers, only 40 units are up for grabs—once they're gone, they're gone. Evolved from the high-degree-of-freedom, full-hand tactile XHAND 1 Pro platform, the XHAND 1 Pro-U underwent a targeted architectural overhaul centered on the "NVIDIA GEAR DexUMI + GR00T 2 + XHAND 1 Pro-U" data loop. The redesign optimizes exoskeleton ergonomics, natural demonstration, tactile signal stability, and physical grasping, converting joint-level correspondence into high-quality, learnable data. Image Source: ROBOTERA ROBOTERA emphasizes that the XHAND 1 Pro-U is the official high-degree-of-freedom five-finger dexterous hand designed for NVIDIA's GEAR DexUMI. Specifically, to ensure consistency across data collection, skill learning, and physical execution, the XHAND 1 Pro-U redesign focuses on four key areas: First, the four fingertips adopt an underactuated linkage design, streamlined into four joint modules for clearer joint coupling relationships. Second, the thumb position has been redesigned, and one degree of freedom removed to improve the fit, naturalness, and comfort of the GEAR DexUMI glove. Third, the palm tactile layout has been optimized. While retaining most tactile sensing on the fingertips and palm, one module prone to dynamic deformation was removed to reduce signal drift and observation noise. Fourth, a palm side-sway degree of freedom was added to accommodate hand deformation during grasping, enhancing adaptability to various object shapes and grip postures. Image Source: ROBOTERA Furthermore, the XHAND 1 Pro-U achieves a one-to-one mechanical correspondence with the NVIDIA GEAR DexUMI data acquisition exoskeleton. Its mechanical structure, contact surfaces, and wrist camera mounting positions align perfectly with the exoskeleton, ensuring that joint movement, fingertip force, and camera perspectives have precise counterparts at the execution end. This partnership is built on the ecosystem framework of NVIDIA's GR00T 2. On November 9 (U.S. time), NVIDIA scientist Jim Fan officially unveiled GR00T 2 at the Conference on Robot Learning (CoRL) and announced a full-stack open-source plan. He noted that GPT-6 Astra isn't about disruption: Astra (System 2) handles understanding and planning, while GR00T 2's World-Action Model, WAM (System 1), manages rapid execution. The stronger System 2 becomes, the more it relies on a high-quality System 1. Yet improving System 1 capabilities is impossible without high-quality data. In traditional teleoperation, operators judge contact indirectly through interfaces, often struggling to perceive resistance and slippage. Moreover, the method ties up the target robot throughout the process. By contrast, NVIDIA's GEAR DexUMI allows users to wear an exoskeleton to manipulate real objects directly. A data collection exoskeleton built for the GR00T 2 ecosystem, GEAR DexUMI preserves natural tactile sensation while recording joint states, fingertip forces, and contact information that correspond to the target dexterous hand. It doesn't occupy the robot, allowing a single operator to generate data continuously in parallel, and enables high-speed natural demonstrations in unstructured environments. This creates a complete pipeline: DexUMI captures human demonstrations, GR00T 2 trains exclusively on this data, and finally, a robot equipped with dual-arm XHAND 1 Pro-U units executes the task autonomously. The entire process requires no target-task teleoperation or reliance on additional robot demonstrations. According to reports, GR00T 2 achieved complete task success rates of 60%, 80%, and 75% in preparing banana yogurt, scooping cereal, and striking a match to light a lamp, respectively. Under the same experimental conditions, GR00T 1.7 and π0.5 failed to complete any of the tasks. Coinciding with the limited launch of the XHAND 1 Pro-U, NVIDIA has open-sourced the GEAR DexUMI acquisition hardware design on the official GR00T 2 project page. The XHAND 1 Pro-U serves as the off-the-shelf execution platform, providing the hardware foundation for cross-embodiment post-training and real-world verification.