Why Is a Century-Old Bearing Maker Betting Its Factories on Two Rival Humanoid Robots?

Three Deals, One Company, Three Different Kinds of Promise

Schaeffler makes bearings and, increasingly, the actuators and gearboxes that let a robot's joint move at all – a business it has quietly built into every major industrial market for a decade. This year it placed three separate bets on humanoid robots, and the three bets do not read the same way at all.

The most concrete one involves no legs. In January, Schaeffler agreed to deploy a four-digit number of wheeled humanoids – outside reporting narrows that to roughly 1,000 to 2,000 – built by the UK startup Humanoid Inc., across its global factories by 2032. This is not a pilot in the loose sense the industry usually means.

It names two German sites: Herzogenaurach, where the robots go straight into box-handling on a live production line, and Schweinfurt, running a six-month sequence – three months of capability testing, three months validating "stable, continuous operation approaching full production scale" – from December 2026 to June 2027. The first machines are due to be working before the end of this year. Schaeffler also becomes Humanoid Inc.'s preferred actuator supplier, covering more than half its demand through 2031.

The second bet walks on two legs, and it is much earlier. In April, Schaeffler committed to putting at least 1,000 of Hexagon's AEON humanoids – a 165-centimeter, 34-degree-of-freedom bipedal platform – into its factories by 2032. Through mid-August, what actually exists of that commitment is a training facility. AEON is being taught tasks there through imitation learning and repeated practice, with no production-floor start date, no unit count for a first wave, and no published success rate for the training itself.

The third bet is Schaeffler betting on itself. It has developed a way to press, rather than machine, the strain-wave gearboxes that sit inside a humanoid's joints – cutting manufacturing cost by more than 25 percent and material use by more than 75 percent, according to the company, with mass production starting in Germany in 2027. Schaeffler says the formed gearboxes reach "comparable torques and efficiencies" to the machined ones it has already shipped more than 2 million of over the past decade. It names no specific humanoid-maker customer for this new process beyond its own two pilots.

Lay the three side by side and a pattern falls out that none of Schaeffler's own announcements states: dates and named sites for the wheeled robot, a gym membership for the bipedal one, and a manufacturing claim with no customer attached for the part supposed to make both possible.

Why the Wheels Got There First

The gap is not a branding accident. Humanoid Inc.'s own case for wheels is a control-engineering argument: on a factory floor doing box handling, tote logistics and container transfer, a robot does not need legs, and a bipedal system spends a real share of its control budget simply staying upright rather than doing the work in front of it. A wheeled base is mechanically simpler, and reporting on the deal also ties it to an easier path toward CE safety certification, which Humanoid Inc. is targeting for 2027. None of that makes bipedal humanoids a mistake.

It says that for the narrow, structured task Schaeffler is paying for first, legs are the harder and slower way to get there – and Schaeffler, spending its own factory floor space on the answer, picked the easier one to move first.

Bipedal locomotion still matters for the tasks a wheeled base genuinely cannot reach – stairs, uneven ground, environments built for human bodies rather than factory aisles. Schaeffler is not choosing between the two forms; it is running both experiments on its own property at once, which is itself informative. A company confident the wheeled bet would win outright would not also be paying to train a second, bipedal robot inside its own walls.

The Number Neither Robot's Paperwork Contains

Here is what the wheeled deal's specificity does not cover: how long the actuator survives. Strain-wave gearboxes – the component family in both Schaeffler's own manufacturing claim and, in some form, most humanoid joints – fail primarily through fatigue fracture of the flexspline, the thin steel cup that flexes on every actuation. That failure mode is driven by cyclic stress, meaning it accumulates with repeated use rather than showing up on a single load test.

"Validation testing concluded" and "comparable torques and efficiencies" – Schaeffler's own language for the new gearbox – describe a part that performs, not one whose service life under a robot's real duty cycle has been published anywhere.

Nor does any safety standard currently require that disclosure. ISO 10218-1, the standard just updated in 2025 to cover this generation of industrial and collaborative robots, scopes itself entirely to eliminating or reducing hazard – inherent safe design, protective measures, information for use. It says nothing about how many cycles a joint should survive before it is retired. A robot can be certified safe on day one without anyone being required to say what happens on day one thousand.

This is not a Schaeffler-specific gap. Boston Dynamics – the humanoid maker that publishes more technical detail than almost anyone in the category – lists Atlas's battery life, payload, IP rating and degrees of freedom on its own product page, and states plainly that Atlas is in "field testing" at a Hyundai facility, not production deployment. There is no uptime figure, no duty-cycle number, no mean-time-between-failure statistic anywhere on that page either. If the most spec-transparent company in humanoid robotics does not publish the number, it is fair to assume nobody currently can.