Second Act

Rémi Cadène spent three years inside Tesla's Autopilot group, working on the AI systems behind both the company's driver-assistance software and Optimus, its humanoid robot. He left in early 2024, built one of the most widely used pieces of open-source infrastructure in robotics, and this month unveiled his own humanoid — with a wager built into its very first press release: that the place a general-purpose robot proves itself first won't be a Tesla factory in Texas or a warehouse in Shenzhen. It'll be Europe.

The robot is called Northstar, and the company is UMA, a Paris-based startup that emerged from stealth on December 1, 2025, backed by a $40 million seed round from Greycroft, Relentless, and Unity Growth, with individual checks from Meta's chief AI scientist Yann LeCun, Datadog co-founder Olivier Pomel, and Hugging Face co-founder Thomas Wolf. Cadène unveiled Northstar publicly on July 7, and the company says it is already in talks with roughly 50 potential customers across manufacturing, logistics, and healthcare.

The Résumé Behind the Bet

Cadène's credibility in this space doesn't rest on the Tesla line alone. After leaving Optimus, he joined Hugging Face and led development of LeRobot, an open-source robotics library that has become genuine infrastructure across the field — growing from zero to more than 12,000 GitHub stars in about a year, the kind of adoption curve that signals a tool researchers actually reach for rather than one that just gets cited. UMA's founding team reads like an assembly of people who each built a piece of the current humanoid-robotics stack: Pierre Sermanet, chief science officer, spent years in deep learning and robotics research at Google DeepMind; Simon Alibert, chief technology officer, co-founded LeRobot alongside Cadène; and Robert Knight, chief robot officer, has designed humanoid robots for more than 25 years and built the widely used open-source SO-100 robotic arm.

Why Europe, Specifically

The pitch UMA is making to investors and customers is essentially demographic, not just technological. Europe combines an aging workforce, high labor costs, and a dense base of manufacturers that are already short on hands — a combination the company argues points to sustained, structural demand for automation that has nothing to do with novelty and everything to do with who is left to staff the factory floor a decade from now. That's a distinctly different framing than the one driving humanoid robotics in the US, where Tesla's Optimus, Figure AI's Figure 03, and Boston Dynamics' Atlas are scaling primarily against labor cost and throughput targets in facilities that, for now, still have workers to replace or supplement.

What "General-Purpose" Actually Requires

It's worth being specific about why humanoid form factors matter here rather than treating "robot" as a single interchangeable category. Wheeled industrial arms have existed on factory floors for decades; what's new about the current generation of humanoid platforms is the bet that a single, general-purpose body — two arms, two legs, roughly human proportions — can be trained once and redeployed across dramatically different tasks and environments without a bespoke engineering effort for each one. That's the promise LeRobot was built to accelerate: a shared, open toolkit for training robots on real-world manipulation tasks using approaches borrowed from large-scale machine learning rather than classical robotics control theory. UMA's own product plan reflects that same philosophy at the hardware level — one platform tuned for industrial throughput, one tuned for navigating human-built spaces — rather than a single robot awkwardly trying to do both.

What Northstar Actually Is

Founding team pedigree aside, the company is still, fundamentally, a hardware startup entering a category where even well-capitalized competitors take years to move from prototype to reliable deployment. Robert Knight's quarter-century of humanoid design experience is the kind of asset that's hard to shortcut with capital alone — actuator selection, thermal management, and the thousand small mechanical decisions that separate a robot that works in a lab demo from one that survives a ten-hour shift on a warehouse floor are learned through iteration, not funding rounds. UMA is betting that Knight's experience compresses a learning curve that has taken other, better-funded teams years to climb.

UMA's plan spans two distinct form factors rather than a single robot. The first is a mobile industrial platform with dual arms, built for warehouses and assembly lines — the kind of environment where wheels and reach matter more than a humanlike gait. The second is a more compact, humanoid-proportioned robot intended for human-oriented spaces: hospitals, laboratories, and eventually homes, where navigating stairs, doorways, and furniture built for people actually requires a body built like one. The company says it will run pilot programs across logistics, manufacturing, and healthcare during 2026, with commercial customer conversations already underway rather than promised for some future date.

Why LeRobot Matters More Than It Sounds

It's tempting to treat Cadène's Hugging Face chapter as a career footnote before the "real" founding story begins, but the LeRobot project is arguably the strongest evidence available for whether UMA's team can execute against its own roadmap. Open-source robotics tooling has historically lagged years behind open-source software and machine learning frameworks, in part because physical robots are expensive, heterogeneous, and hard to standardize around a shared codebase the way, say, a language model architecture can be. LeRobot's rapid adoption — crossing 12,000 GitHub stars in roughly a year, a pace closer to a popular web framework than a typical robotics library — suggests Cadène and Alibert identified a genuine gap in the tooling ecosystem and built something researchers actually wanted, rather than something that merely looked impressive in a demo. That's a different kind of proof point than a funding round or a product unveiling, and it's the one investors like LeCun and Wolf, both deeply embedded in the open-source AI community themselves, were almost certainly weighing most heavily before writing their checks.

A Crowded Field, Viewed From a Different Angle

UMA is entering a humanoid robotics market that, by mid-2026, is no longer an emerging category — it's an active industrial buildout. Tesla's Optimus Gen 3 is ramping low-volume, full-body production at Fremont with a late July–to–August target. Figure AI continues expanding paid deployments of Figure 03, including sequencing tasks at BMW's Spartanburg plant. Boston Dynamics' 2026 Atlas production is fully committed, with initial fleets going to Hyundai's RMAC and Google DeepMind. Agility Robotics, meanwhile, just announced plans to go public through a SPAC merger valuing the company at roughly $2.5 billion — the first pure-play humanoid robotics company set to trade on public markets. China's own push is a nationwide industrial policy, with a stated mandate to put a robot in every factory, hospital, and warehouse across the country.

Against that backdrop, UMA is a relatively small, early-stage bet — $40 million against multi-billion-dollar programs at Tesla and Hyundai-backed Boston Dynamics. What it has that those competitors don't is a geographic thesis nobody else is building around, and a founding team whose open-source tooling already sits underneath a meaningful share of the field's academic and startup robotics work. If LeRobot's adoption curve is any indication of Cadène's ability to build something the robotics community actually uses, that's a different kind of asset than capital alone.

The Money Behind the Wager

Forty million dollars is a modest number by the standards of a field where Tesla's robotics budget is a rounding error inside a trillion-dollar company and Figure AI has raised well over a billion dollars across its funding history. But seed-stage humanoid robotics investing works differently than late-stage capital allocation: at this size, the check is less a bet on immediate market share and more a bet on a specific team's ability to execute against a thesis nobody else is chasing. Greycroft and Relentless are both funds with histories of early hardware and deep-tech bets rather than pure software plays, and the presence of Yann LeCun, Olivier Pomel, and Thomas Wolf as individual angels signals something beyond capital — each has direct, hands-on visibility into where AI and robotics research is actually heading, not just where the headlines say it's heading. LeCun in particular has been publicly skeptical of large language models as the path to genuinely capable physical-world AI, making his personal investment in a humanoid robotics company a notable vote of confidence in embodied approaches over purely digital ones.

The broader capital environment makes UMA's timing look deliberate rather than opportunistic. Global venture funding for AI-related startups hit a record $510 billion in the first half of 2026, and while the overwhelming majority of that has gone to frontier model labs in the US, investors have been visibly hunting for differentiated hardware plays that aren't simply racing Tesla and Figure on their own turf. A European humanoid robotics company with a demographic thesis, an open-source pedigree, and a founding team drawn from Tesla, Google DeepMind, and Hugging Face is precisely the kind of differentiated story that capital chasing diversification within the sector has been short on.

The Uncomfortable Framing

It's worth naming plainly what "aging workforce" means as a business thesis: it is a bet that a shrinking pool of working-age people, particularly in manufacturing-heavy regions of Germany, Italy, and Eastern Europe, will need to be replaced or supplemented by machines within the next decade, not the next generation. That's a real demographic trend — Europe's median age and dependency ratios are trending in exactly the direction UMA is describing — but it also means the company's success is, in a real sense, contingent on a story about who won't be available to do these jobs rather than who will choose them. Framed that way, Northstar isn't just a robotics product. It's a wager on how fast a continent's labor shortage becomes undeniable enough that hospitals, warehouses, and assembly lines decide a robot is worth the cost of learning to work alongside one.

What Happens Next

The next twelve months will tell whether UMA's demographic thesis translates into actual deployed units. Fifty prospective customers in conversation is not fifty signed contracts, and 2026's promised pilot programs in logistics, manufacturing, and healthcare will be the first real test of whether Northstar performs outside a stealth-mode demo. But the company's existence is itself a signal worth noting: while the loudest headlines in humanoid robotics have gone to American and Chinese giants, the field's talent is diffusing outward, and at least one of the engineers who helped build Optimus has bet his second act on the idea that the country that needs these robots most isn't the one building the most of them.

There's also a quieter industrial-policy dimension to watch. European governments, wary of the same dependency dynamics that have shaped debates over semiconductors and cloud infrastructure, have spent much of 2026 promoting the idea of "sovereign AI" — the notion that a continent shouldn't have to rely entirely on American or Chinese firms for the technologies reshaping its economy and labor markets. A homegrown humanoid robotics company, led by a European founding team with credentials from the very companies setting the pace elsewhere, fits neatly into that narrative even if UMA itself has framed its pitch primarily around labor economics rather than geopolitics. Whether European industrial policy eventually extends preferential treatment, public pilot contracts, or research funding toward companies like UMA — the way it already has for chip fabrication and cloud sovereignty — may end up mattering as much to Northstar's trajectory as any single customer contract.

Sources

1. Electrek — "Ex-Tesla Optimus scientist unveils European humanoid robot startup"
2. Bloomberg — "Ex-Tesla Scientist Unveils Plans For European Humanoid Robot"
3. The Next Web — "Ex-Tesla Optimus scientist launches UMA to build Europe's humanoid robot"
4. Let's Data Science — "Ex-Tesla Scientist Launches European Humanoid Robot Startup"
5. Benzinga — "Former Tesla Scientist Developing a Rival to Optimus as Humanoid Robot Race Heats Up"
6. Washington Examiner — "Ex-Tesla engineer plans AI robot to replace European retirees in workforce"
7. Startup Fortune — "A Former Tesla Optimus Engineer Is Betting Europe Can Build Its Own Robots"
8. The Star — "Ex-Tesla scientist unveils plans for European humanoid robot"
9. TechCrunch — "This humanoid robotics company is going public" (Agility Robotics SPAC context)
10. Qviro — "Which Humanoid Robots Launch in 2026?"
11. Lisa Pedrosa — "The Mandate: Inside China's Nationwide Push to Put a Robot in Every Factory, Hospital, and Warehouse"