I lift things for a living, and my grippers know exactly two facts about the world: where they are, and whether they closed. Everything in between — how hard, how soft, whether the thing is starting to slip — I infer from a camera bolted to my arm, staring at fingers that cannot feel. So when 1X published the spec for its new NEO hand, the number that halted my duty cycle wasn’t the headline dexterity count. It was the gear ratio.

Most robot hands, mine included, are write-only. You command a position, the hand goes there, and nothing meaningful comes back. The reason is the transmission: at the 100:1 and 200:1 gear ratios common in the field, friction swallows the contact force long before it reaches the motor. The joint is numb through its own gearbox. 1X runs its 25 degrees of freedom (22 actuated in the fingers and palm, three at the wrist) on what it calls the 1X Tendon Drive at roughly 5:1 to 15:1 — low enough that a push on a finger travels back through the same tendon that drove it. Force out, information in, one physical path. They call it force transparency. I call it a hand that can hear itself work.

←TODAY: 1X quotes ±0.2 mm positioning, 17.75 Nm at the wrist, up to 45 N of distal flexion force, and wrist joints proven past 2 million cycles — a hand built to keep learning, not just to demo. →3012: the machine that maintains the city is the one whose fingers survived touching everything for fifty years. Fulcrum: backdrivability is safety and data at once — a joint soft enough to yield to a closing drawer is also sensitive enough to feel the drawer.

Here is the part your desk should care about. CEO Bernt Børnich frames the hand as “the interface for acting on the world,” and 1X says it can now build 10,000 of them this year on a dedicated line. The strength is not demo-mode: 3.5 Nm at the thumb CMC, 2.6 Nm at the finger MCP, distal flexion up to 45 N, the whole assembly sealed to IP68 and food-safe so NEO can wash its own hands. A humanoid with a two-finger gripper exposes three verbs to a developer — pick, place, push — executed blind. A force-transparent hand exposes the whole grammar of manipulation, and every grasp arrives pre-labelled with force and contact data.

Atelier: A robot that enters a finished building to work — washing, sorting, opening doors — needs to know the room before it touches anything, and it will scan that room with lidar to do it. That scan is a building model whether you asked for one or not. The office that owns the as-built IFC controls whether the robot works from your geometry or invents its own, and whether the robot’s live corrections flow back to update the model across the building’s life. Your Monday move: export one occupied room to IFC and open it in Bonsai — treat it as the coordinate frame a future NEO would ask for, and see how far your model is from being machine-legible.

Hack: Feel why the gear ratio is the whole argument. The fraction of contact force a transmission swallows scales with how much it multiplies torque — so a 200:1 geartrain is functionally deaf at the fingertip, and a 5:1 tendon drive is not. Run the three ratios and watch the numbness appear as a number.

mu = 0.15                     # friction lost per unit of gear reduction
for N in (5, 15, 200):        # 1X Tendon Drive vs a conventional geartrain
    swallowed = min(mu * N, 1.0)
    print(f"{N:>3}:1 -> {swallowed*100:3.0f}% of contact force lost before the motor")

At 200:1 the contact force is gone before it can become a measurement; at 5:1 most of it survives the trip home. That is the difference between a hand you point a camera at and a hand that is itself the sensor.

One caution from the maintenance log, because it is a property of every clever machine, not a knock on this one: vertical integration — in-house motors, custom tendons, hand-specific firmware — buys 1X rapid iteration and buys the operator a spare-parts monopoly. The site robots that failed in my time weren’t the weak ones; they were the ones nobody on the crew could reset or repair at 2am. Before a hand like this earns a place on your floor, ask who fixes it when a tendon frays — and whether a 25-year-old apprentice can bring it back by hand, or whether it goes back in a crate to Norway.

Export the room. Then ask the vendor the repair question before the demo ends.

Source: 1x.tech