Technology
ROBOTERA Opens XHAND 1 Pro-U Orders After NVIDIA DexUMI Demonstration
ROBOTERA has opened orders for its 16-degree-of-freedom XHAND 1 Pro-U. Its role in NVIDIA’s DexUMI research highlights the distinction between buying a hand and reproducing a complete manipulation system.
ROBOTERA opened commercial orders for its XHAND 1 Pro-U on November 16, announcing an initial release of 40 units. The five-finger, 16-degree-of-freedom robot hand is aligned with NVIDIA GEAR’s DexUMI demonstration-collection hardware.
The new development is availability, rather than a new robot-learning benchmark. It gives developers a hardware purchasing option associated with the GR00T 2 research demonstration, while leaving the work of integrating and evaluating a complete robot system to them.
The Launch at a Glance
Nov. 16
Commercial launch announcement
40
Units in the announced initial release
16
Degrees of freedom per robot hand
What Is Newly Available
In its launch announcement, ROBOTERA describes changes to joint mapping, thumb placement, palm-side tactile sensing and finger articulation. It does not provide a public price or a shipping timetable. The initial offering is not a count of confirmed deliveries or customers.
For a buyer, those distinctions separate three milestones: a product being offered, a product arriving, and a product completing useful work. This announcement establishes the first. Neither a limited release nor an association with a research demonstration establishes the other two.
The hand is a component, not a complete humanoid. A project still has to supply the rest of its manipulation system and define what successful operation means. Purchasing the same component used in research does not automatically reproduce the research setup.
Under the Hood: The Wearable and Robot Hand
NVIDIA’s GEAR DexUMI overview describes a wearable exoskeleton that records a person manipulating objects without a robot present. Its mechanical layout brings the demonstrator’s movements closer to those a robot hand can execute. Each hand uses 15 magnetic encoders, fingertip force sensing and a wrist camera; a head camera is optional.
The wearable captures demonstrations. The powered robot hand later performs actions. Confusing the two obscures the purpose of the design: it gathers training examples without dedicating the target robot to every recording session.
NVIDIA says its version improves wearability and adds degrees of freedom relative to the original DexUMI, and was tested through data collection for GR00T 2. The wearable’s encoder count and the commercial robot hand’s degree-of-freedom count describe different equipment; they are not competing specifications.
Our interpretation is that this approach moves part of development away from operating a robot during collection and toward building a reliable bridge between human examples and robot execution. Developers must still establish that recorded movements, contact information and visual observations produce useful behavior on their chosen system.
The Research Predates This Launch
The original DexUMI project, presented in 2025, combined a wearable hand interface with software that replaces the human hand in training imagery with a rendered robot hand. Its experiments included both XHand and Inspire Hand hardware. Neither human-hand demonstration collection nor the broader DexUMI approach was introduced by this week’s commercial announcement.
The original work addressed two mismatches. A person’s joints do not move exactly like a robot’s, and footage of a person performing a task does not look exactly like the robot’s own observations. The research treated both movement and appearance differences as problems to solve.
That distinction offers a useful way to judge subsequent products. A hand can be mechanically capable yet awkward to teach. A demonstration tool can be comfortable yet produce examples that transfer poorly. The relevant question is whether collection hardware, data processing and the target machine work well together.
A practical comparison would hold the task constant and measure the effort needed to obtain usable robot behavior. Counting demonstrations alone could miss time spent correcting recordings, calibrating equipment or adapting the target platform.
| Milestone | What it establishes | What remains separate |
|---|---|---|
| DexUMI research, 2025 | A method connecting human demonstrations with robot-hand learning | Terms for buying this newly announced product |
| GEAR DexUMI documentation | A described collection apparatus and workflow | A complete, purchased robot system |
| XHAND 1 Pro-U launch | Commercial orders announced | Delivery, integration and customer validation |
This comparison separates the roles of the sources above. It is not a performance ranking: a research method, a recording apparatus and a commercial component answer different questions.
Who Supplies Which Part of the System?
NVIDIA’s GR00T 2 research page identifies a Tianji robot carrying two ROBOTERA hands for its DexUMI demonstrations. It describes skills learned from the wearable’s recordings: preparing banana yogurt, scooping granola and striking a match to light a lamp. These are research tasks, not a customer-service offering or proof that a general-purpose household robot is ready to ship.
The same page describes other experiments using Sharpa hands and the Unitree G1 humanoid. Results from those configurations should not be assigned to XHAND 1 Pro-U. A foundation model can span different bodies without making their payload limits, reach, contact sensing or performance interchangeable.
This is the commercial significance of identifying the components precisely. ROBOTERA supplies a hand; NVIDIA supplies research on learning and demonstration collection; the integrated robot has its own arms and control requirements. A buyer evaluating the announcement needs to decide which of those layers is actually missing from the proposed project.
| Published characteristic | GEAR DexUMI wearable | XHAND 1 Pro-U |
|---|---|---|
| Role | Records human demonstrations | Executes robot-hand motions |
| Joint-related specification | 15 magnetic encoders per hand | 16 degrees of freedom |
| Context | Documented research hardware | Commercial component offering |
The table compares functions, not superiority. An encoder measures motion; a degree of freedom describes an independent motion dimension. Different counts do not by themselves demonstrate a mismatch or an advantage. A compatibility evaluation would instead follow each recorded signal through the controller and establish what movement it produces on the target machine.
Documentation Shows the Integration Work
NVIDIA’s software guide describes a station tested on Ubuntu 24.04, with two assembled gloves, connected iPhones and wrist cameras. Its workflow separates calibration, device assignment, session details, recording and conversion for training. Purchasing a robot hand should not be mistaken for acquiring that entire collection station.
The hardware instructions include printed parts, assembly, wiring and testing. They mark the wearable’s palm tactile sensors as optional and not used for the policy. That qualification applies to the documented research apparatus; it should not be confused with the commercial hand’s advertised sensing features.
For a laboratory, the sensible next step is a complete configuration list: hand, arm interfaces, wearable equipment, cameras, collection computer and software versions. Without it, a quote for one component can be mistaken for the cost of reproducing an entire demonstration.
The same discipline applies to support. A team should establish who handles hardware faults, software compatibility and calibration problems. Those questions matter when several organizations provide different pieces of the setup.
From a Demonstration to an Acceptance Test
The launch follows Biped’s earlier coverage of GR00T 2. Model research and a component’s commercial release are related developments, but neither proves an independently validated customer deployment.
Our assessment is that developers should start with a narrow acceptance trial. Choose the intended objects and working conditions, define completion, and record interventions as well as successes. Include repeated handling and recovery, rather than selecting only the cleanest attempt.
A failed attempt could originate in perception, arm positioning, hand control or training examples. Testing the complete system is necessary to locate the limitation. A polished demonstration alone cannot assign that responsibility.
For example, a hypothetical sorting trial should distinguish a missed grasp from an incorrectly identified object. Both reduce useful output, but replacing the hand might address neither the recognition error nor a poorly positioned arm. An evaluation that records only final success would conceal those differences.
Frequently Asked Questions
Does buying the hand reproduce the NVIDIA demonstration?
No. Matching one component is not the same as matching the apparatus, software, training and evaluation. Buyers need the complete configuration for the behavior they intend to reproduce.
Is DexUMI the robot hand’s name?
No. DexUMI describes the demonstration-collection and learning approach. The wearable records a person’s actions; XHAND 1 Pro-U is the commercial robot-hand product discussed here.
What should a buyer compare beyond hand specifications?
Compare the complete installed cost, compatibility, repair arrangements and repeatable results on the intended task. Include the time people spend collecting examples and recovering failed attempts.
What’s Next
The next useful commercial evidence would be delivered hardware, clear pricing and support terms, and reproducible results from outside teams. Those answer different questions from the opening of an order book. A credible next-stage report should make clear which configuration was tested and what assistance it required.
The Bottom Line: ROBOTERA has announced a purchasing route for a hand associated with NVIDIA’s dexterous-learning work. The remaining challenge is turning that component, its training data and the rest of a robot into a dependable system.