The race to build a useful humanoid robot may ultimately be decided not by how fast it walks or how naturally it speaks, but by what it can do with its hands.
On July 9, 2026, 1X Technologies unveiled a new generation of hands for its 1X NEO home humanoid. The system provides 25 force-controlled degrees of freedom, tactile sensing across the fingers and palm, and a tendon-driven architecture designed to approach the movement, sensitivity and strength of a human hand.
According to 1X’s technical announcement, the hands will be installed on every production NEO. The company argues that the new hardware removes a fundamental limitation that has prevented humanoid robots from reliably interacting with the objects, tools and environments built for people.
The result is one of the most ambitious robotic-hand systems yet presented for a consumer-focused humanoid. It is also an important test of whether advanced manipulation hardware can be translated from carefully produced demonstrations into dependable, autonomous work inside real homes.
A robotic hand designed to feel, not just move
Degrees of freedom, usually shortened to DoF, describe the independent ways in which a mechanical system can move. A simple two-finger gripper may only open, close and rotate. A human hand is commonly modelled with approximately 27 degrees of freedom, allowing the fingers and wrist to combine thousands of possible poses and grasping strategies.
NEO’s new system has 22 actively controlled degrees of freedom across the fingers and palm, plus three at the wrist. All 25 are force-controlled and backdrivable, meaning that the robot can both apply force and detect force returning through the same mechanical system.
Instead of placing heavy motors inside the fingers, 1X locates the motors in NEO’s forearms. Those motors pull proprietary tendon-like cables through the wrist, recreating some of the mechanical principles used by muscles and tendons in the human arm.
The transmission operates at comparatively low gear ratios of around 5:1 to 15:1. This allows the fingers to yield when pushed rather than remaining rigid, while enabling the robot to estimate how much force is being applied at each joint. 1X calls this capability “force transparency.”
The hands also include tactile sensing across their surfaces. The sensing system measures contact location, pressure and shear—the sideways force that can indicate when an object is beginning to slip.
That feedback could allow NEO to adjust its grip before a glass, tool or piece of food falls from its hand. It also gives the robot information that cameras alone cannot reliably provide when handling transparent, deformable, partially hidden or unusually shaped objects.
What can NEO do with its new hands?
In its launch demonstrations, 1X showed NEO performing tasks that require very different combinations of strength, precision and coordination.
The robot assembled LEGO pieces, picked up individual screws and coins, installed a light bulb, operated a screwdriver, zipped a jacket, sorted grapes by colour, poured tea, caught a soft ball and connected a USB-C cable. It was also shown holding a wine glass, cleaning a surface and communicating through sign language.
These examples are significant because they extend beyond conventional robotic picking. Connecting a cable or operating a zip requires alignment, controlled force and continuous correction. Handling grapes, glasses and origami requires the robot to apply enough pressure to maintain its grip without damaging the object.
The fingers can also hyperextend beyond the normal range of a human hand and move rapidly enough to perform high-speed finger-drumming motions. According to 1X, the system achieves positioning accuracy of approximately ±0.2 millimetres.
The headline specifications include:
- 25 force-controlled degrees of freedom: 22 across the fingers and palm, plus three at the wrist.
- Tendon-driven actuation: Motors in the forearm pull cables connected to the fingers.
- Tactile and shear sensing: The hand can detect pressure, contact position and object slippage.
- IP68 protection: The sealed hands are designed to withstand dust and immersion in water.
- Food-safe materials: NEO can handle food and wash its own hands.
- Up to 45 newtons of distal finger-flexion force.
- 17.75 Nm of wrist torque.
- Approximately ±0.2 mm positioning accuracy.
1X also subjected the hands to impact demonstrations involving drawers, direct strikes and even a hammer. Because the fingers are backdrivable and have relatively low distal mass, they can move with an external impact rather than resisting it as a rigid mechanism would.
Why humanoid hands are so difficult to build
Robotic manipulation traditionally involves a compromise.
Industrial grippers can be strong, fast and reliable, but they are normally designed for a narrow range of objects and repeatable conditions. More anthropomorphic hands can perform a wider variety of movements, but they often become expensive, heavy, fragile or difficult to manufacture and maintain.
A useful humanoid hand must combine qualities that frequently conflict with one another. It needs enough strength to lift bags and operate tools, but enough sensitivity to hold glassware. It must be compact, but contain multiple actuators, transmissions, sensors and electrical connections. It must also survive thousands of daily interactions without losing calibration or breaking tendons.
1X says the components and finger assemblies have undergone millions of test cycles, while the wrist joints have been tested beyond two million cycles under high loads. The company has also designed a dedicated production line for the hands at its NEO Factory in Hayward, California.
Manufacturability may prove as important as dexterity. A technically impressive hand has limited commercial value if it requires specialist assembly, frequent repairs or components that cannot be produced at scale.
Why the hands matter for robot intelligence
1X describes the hand as an “API to the physical world.” The analogy is useful: software determines what a robot intends to do, but the hands determine which physical actions are actually available.
A robot equipped with a basic gripper can be trained to pick up a box. A sufficiently dexterous five-finger hand could potentially operate the same tools, switches, handles, appliances and containers that people already use.
This matters particularly inside homes. Domestic environments are not organised around standardised pallets or identical factory components. They contain clothing, cables, utensils, packaging, food, fragile decorations and objects that regularly change position.
A human-like hand may also make it easier to train humanoid robots from human demonstrations. Movements captured from a person folding clothes or preparing food are more directly transferable when the robot has a similar number of fingers, comparable joints and a genuinely opposable thumb.
Every interaction can then generate additional training data. Joint position reveals how the hand moved, motor feedback estimates the forces involved, and tactile sensors record where contact occurred. In principle, that creates a continuous loop in which the robot acts, feels the result and improves its future behaviour.
However, the hardware does not solve the intelligence problem by itself. NEO still requires perception, planning and control models capable of selecting the correct movement and recovering when an object moves, deforms or behaves unexpectedly.
Image Credits: 1X TechThe demonstrations do not prove full autonomy
The launch footage establishes an impressive upper limit for the physical hardware, but it should not be interpreted as evidence that NEO can autonomously complete every demonstrated task inside an unfamiliar home.
1X confirmed to independent reporters that the published demonstrations included a mixture of autonomous operation and human-controlled movement. Some sequences were generated by the robot itself, while others used an operator to demonstrate the maximum capability of the hands.
The sign-language demonstration was identified as autonomous, but 1X has not publicly classified every individual sequence in the launch film.
This distinction is important. Teleoperation can prove that the mechanical system is capable of producing a movement. Autonomous manipulation requires the robot to perceive the situation, choose an appropriate grasp, control the movement and recover from errors without a person directly guiding it.
NEO already includes an “Expert Mode” through which a remote 1X operator can guide the robot when it encounters an unfamiliar chore. The company presents this as both a service for owners and a way of collecting demonstrations that can later be used to train autonomous behaviour.
For buyers evaluating the 1X NEO, the practical question is therefore not simply whether its hands can perform a task. It is how frequently the complete robot can perform that task autonomously, in a real environment, without intervention.
Built for the home—and potentially for mass production
NEO is positioned as a home humanoid rather than an industrial robot. It is designed to perform domestic chores, retrieve and organise objects, interact through speech and gradually acquire new capabilities through software updates and real-world training.
1X currently lists an Early Access price of $20,000, with priority delivery in the United States during 2026. A subscription option is advertised at $499 per month, with broader international expansion planned from 2027. Buyers can review its current specifications, commercial positioning and availability through the Anton Robots 1X NEO profile.
The company says its 58,000-square-foot Hayward facility has begun full-scale NEO production and currently has capacity for up to 10,000 robots per year. With additional automation and a second California facility, 1X is targeting annual production of more than 100,000 units by the end of 2027.
Those targets remain ambitious, but the decision to manufacture the motors, tendons, structural components, electronics, tactile skin and hand assemblies internally gives 1X tighter control over both product development and production.
What this means for the humanoid robot race
Walking demonstrations once dominated the humanoid industry. The competitive focus is now shifting toward manipulation: whether robots can reliably interact with the enormous variety of objects encountered in homes and workplaces.
The new NEO hands give 1X a credible hardware platform for that challenge. Their combination of human-scale movement, force feedback, tactile sensing, waterproofing and manufacturability addresses many of the weaknesses that have historically limited dexterous robotic hands.
But 1X’s strongest claims—particularly that the hands match or exceed human performance—still come primarily from the company’s own measurements and demonstrations. Publicly available evidence does not yet constitute a standardised, independent benchmark of long-term autonomous performance.
The next milestone will not be another video showing what the fingers can do under ideal conditions. It will be evidence that production NEO robots can use those fingers reliably across hundreds of ordinary tasks, objects and homes.
For now, the announcement represents a meaningful hardware advance and one of the clearest indications yet that humanoid developers are moving beyond locomotion and focusing on the part of the robot that performs the actual work.
Readers can explore other commercially relevant humanoid robots or compare robot models side by side through Anton Robots.
The bottom line
1X has built more than an articulated robotic gripper. NEO’s new hands combine movement, force control and tactile perception in a system intended to operate at human scale and eventually be manufactured in significant volumes.
If the company can pair that hardware with sufficiently reliable autonomous control, NEO could manipulate many of the everyday objects that have remained out of reach for conventional robots.
The hands appear ready to perform. The unanswered question is how quickly NEO’s intelligence can learn to use them without human help.