Ex-Tesla Scientist Launches UMA to Build Europe's Humanoid Robot Rival to Optimus

Ex-Tesla Scientist Takes on European Humanoid Robot Challenge

Rémi Cadène spent roughly three years at Tesla, from 2021 to 2024, working on the AI behind Autopilot and building the first neural networks for Tesla Optimus 1

1

. Now the ex-Tesla scientist is launching his own venture from Paris, aiming to prove that Europe can compete in the humanoid robot race against American and Chinese rivals. UMA, short for Universal Mechanical Assistant, emerged from stealth in December 2025 with plans for a lightweight, AI-powered humanoid robot called Northstar 2

2

. The Paris-based startup is targeting manufacturing plants, logistics warehouses, and eventually homes, with Europe as its beachhead market rather than the United States or Asia.

Real-Time Learning Sets Northstar Apart

The Northstar robot weighs 40kg and is being designed with a distinctive capability: real-time learning through observation 4

4

. Unlike many industrial robots that require specific programming for each new task, Northstar watches a human completing a task, attempts to replicate it, and gradually improves its performance through repetition. Cadène has compared this approach to how children learn to tie their shoelaces by watching, practicing, and refining the skill. Demos at the company's Paris lab show robotic arms using computer vision to sort objects by color. UMA says it builds and owns its entire stack, from hardware to software, which allows for end-to-end control and safety across every level 4

4

.

Supergroup Team and High-Profile Backing

Cadène has assembled a founding team with credentials spanning the biggest names in AI robotics. The team includes chief science officer Pierre Sermanet, a veteran of Google DeepMind and NYU, chief technology officer Simon Alibert, a LeRobot co-founder from Hugging Face, and chief robot officer Robert Knight, the designer behind the widely used open-source SO-100 arm 1

1

. Before founding UMA, Cadène joined Hugging Face in early 2024 to lead LeRobot, the open-source toolkit that became core infrastructure for robot learning worldwide, growing from zero to more than 12,000 GitHub stars in about a year 2

2

. The investor list reads like a who's who of the field, with backers including Greycroft, Red River West, Kima Ventures, and Factorial 1

1

. Advisers include Meta's chief AI scientist and Turing Award winner Yann LeCun, alongside Hugging Face co-founder Thomas Wolf 1

1

. Before the public launch, Cadène was reported to be seeking around $40mn in seed funding, though UMA has not confirmed a final figure 1

1

.

Europe-First Strategy Targets Labor Shortages

UMA's strategy leans heavily on Europe's acute labor shortages, from warehouses with punishing staff turnover to healthcare systems short of millions of workers 1

1

. "Labor costs are very high and, given the demographic trends, there will be significant demand," Cadène said, pointing to the continent's aging workforce and dense industrial base 2

2

. The company says it is already in conversations with about 50 potential customers about how they might put Northstar to work 3

3

. UMA plans to run several industrial pilot programs in logistics, manufacturing, and healthcare during 2026, positioning Northstar as a machine that can perceive, move, and manipulate objects in messy real settings rather than stage-managed demos 1

1

.

Crowded Field with Figure AI Leading Deployments

The humanoid robot race has become intensely competitive, with multiple players pushing toward commercial deployment. Figure AI appears furthest ahead on actual deployment, with its robots running shifts at BMW's Spartanburg plant, where the company says they helped build tens of thousands of vehicles 2

2

. Boston Dynamics is now working with Hyundai on a similar collaboration 4

4

. Tesla is targeting low-volume production of its third-generation Optimus at Fremont this summer, though CEO Elon Musk admitted in January that no Optimus robots are doing "useful work" at Tesla in any material way, and the company has announced no external customers 2

2

. European peers such as Germany's Neura Robotics and Stuttgart-based Sereact have raised heavily over the past year 1

1

. UMA's wager is that a European team with open-source roots can move faster on the software than on the hardware, and that customers on the continent would rather buy a robot built closer to home 1

1

. Based on images shared so far, the project appears far behind on the robotics front, but the biggest bottleneck with humanoid robots appears to be software: having the robots understand and interact intelligently and autonomously with their physical environment 2

2

.