The quest for humanoid dexterity is heating up. With Wuji's Hand 2 spec and Tesla Optimus V3 hand and arm patents, we have two diverging but ambitious bets on the future of the synthetic tendon. Unfortunately, one of them did not survive even one week.

The CEO of Wuji Technology, Yunzhe Pan’s business pivot and engineering position have both led to his rejection of the biological tendon. To Pan, the tendon is a mechanical dead-end defined by "creep"—the inevitable loosening that destroys long-term reliability. He argues that matching the human 22-DOF benchmark with a dual-tendon architecture would necessitate a staggering 44 motors, creating a maintenance nightmare. Even spring-based systems are dismissed as shortcuts, as non-linear spring forces are notoriously difficult to model accurately. Pan insists that making tendons reliable requires an upstream ecosystem of materials and modeling that currently does not exist. For Wuji Technology, the only path to a stable machine involves removing the part, not hiding it.

Tesla’s dexterity strategy, detailed in their April 2026 Optimus V3 patents, represents a massive bet on relocating rather than deleting complexity. While Wuji embeds actuators in the finger segments, Tesla moves all actuators into the forearm, threading three tendons per finger through the wrist to achieve human-level DOF*. There is a cold logic here: the hand must remain slim and lightweight to be useful, and heat is the relentless enemy of precision. By treating the tendon as a necessary messenger, Tesla avoids the "fat finger" syndrome in direct-drive designs. That wrist-routed mechanism, the rolling-contact joint meant to thread three tendons cleanly through a moving wrist, is precisely the kind of system Pan is critical about. On April 19th, three days after the patent published, Musk confirmed it: 'This one didn't actually work.'

This isn't a contest with a winner. Both bets remain unproven. Wuji's Hand 2 specs are explicitly marked Beta1 and subject to change, and Tesla's own targets cap out at low-volume production later this year. Neither has shown its hand can survive a factory floor without someone standing by to fix it. But only one of them has had its three-day reckoning. Wuji hasn't yet faced the moment where its own founder has to admit a core mechanism doesn't work in practice — that's still ahead of it, if it's coming at all. Tesla just lived through that moment in public, and moved on like it was routine. If a company with Tesla's resources can publish a patent and disown it before the week is out, the real question isn't which architecture wins. It's how much of what either company says about its hand today will still be true by the time either one actually ships.

^{*DOF is a count of the independent ways a each joint can bend, rotate, or pivot. For reference, a human hand has 22 DOF.}

Within a ten-week window, Bosch unveiled its robotics strategy at an event in Berlin, Schaeffler accepted an award for its actuators, and Valeo announced a humanoid training center in Nanjing.

Bosch's robotics strategy centered on MEMS (tiny sensors that act as a robot's sense of touch) at the same Berlin event where it disclosed a €363 million post-tax loss, its first since 2009. CEO Stefan Hartung illustrated the scale of the bet with a hypothetical: if robots needed as many touch sensors as the 4 million found in human skin, four years of the world's entire sensor production would equip just 12,500 robots. He called this proof of "immense potential", even as Bosch's core business battles weak demand.

Schaeffler accepted the Hermes innovation award for its actuators, the mechanical motors that allow robot joints to move, before German Chancellor Friedrich Merz. Schaeffler puts the actuator share of total humanoid costs at roughly 50 percent. This figure however comes from Schaeffler, untested against any actual order book.

Valeo announced a humanoid training center in Nanjing in language it calls "China Speed" — three sentences buried at the end of a 1,500-word release about electric vehicles. Three companies, independently, reached for the same story at the same time.

Germany's own trade-investment agency is amplifying that story. Germany Trade and Invest, the federal body tasked with attracting foreign business, frames humanoid robotics as the country's next industrial advantage, citing analyst forecasts of $80 billion to $150 billion in annual global sales by 2035. A range with a $70 billion spread isn't a forecast so much as a permission slip that is wide enough to justify almost any claim made underneath it.

The agency's own proof point is a single Apollo robot, built by the Texas-based startup Apptronik, performing logistics tasks at a Mercedes-Benz training center in Berlin. Assembly-line work remains, in the agency's words, a long-term goal. Germany's flagship case for its own humanoid future currently runs on someone else's robot.

Across Bosch, Schaeffler, and Valeo, there is no named customer placing a volume order. No start-of-production date. No buyer who has signed anything. Three companies and one government agency are telling the same story about a market that, on the evidence so far, doesn't have a single paying customer in it.

A reckoning lands twice in this issue. One robotics giant abandons its own patent in public, three days after publishing it. Three distressed automotive suppliers, independently, reach for the same uncertain bet within ten weeks of each other. Both stories turn on the same uncomfortable question: how much of what's being announced is actually load-bearing.