UMA's Northstar humanoid robot runs on open-source AI from ex-Tesla Optimus scientist

A significant shift in the global humanoid robotics race emerged this week as Rémi Cadène, a former Tesla scientist who helped pioneer the AI underpinnings of Elon Musk's Optimus robot, unveiled his own Paris-based startup, UMA, and its ambitious humanoid robot project, Northstar. The announcement is more than a routine product launch; it represents a strategic play for European independence in a field dominated by American and Chinese tech giants. Cadène's deep expertise—three years at Tesla building neural networks for Autopilot and the first iteration of Optimus's AI, followed by his leadership of Hugging Face's open-source LeRobot toolkit—gives UMA immediate credibility. The startup is already in conversations with approximately 50 potential customers across manufacturing, logistics, and home sectors, signaling robust early demand for humanoid labor even before a prototype is publicly demonstrated. While funding amounts have not been disclosed, UMA emerged from stealth in December 2025 with an investor roster that includes Greycroft, Red River West, Kima Ventures, and Factorial, alongside heavyweight advisors like Meta's chief AI scientist Yann LeCun and Hugging Face co-founder Thomas Wolf. This backing positions UMA as one of Europe's most credible bets in a market projected to be worth hundreds of billions by the 2030s. The move also underscores a trend of critical talent dispersing from Tesla's Optimus program: Cadène is not the first alumnus to leave to build rival systems, pointing to both the program's productivity and its retention challenges. For Europe, which has struggled to produce a global hardware champion, the focus on 'Europe-first' deployment may prove politically and economically savvy, potentially translating into regulatory and partnership advantages. However, the path to production is fraught; humanoid robots remain expensive and technically immature compared to stationary manipulators, and UMA must convert customer conversations into binding agreements, then navigate the gulf between an AI-first prototype and a reliable, safe product. Cadène's open-source legacy via LeRobot could be a differentiator, enabling faster community-driven iteration, but it might also expose competitive trade secrets. Ultimately, UMA's success will hinge on execution speed and the ability to integrate lightweight, affordable hardware with the advanced AI systems Cadène and his team—which includes DeepMind veteran Pierre Sermanet, LeRobot co-founder Simon Alibert, and open-source robotic arm designer Robert Knight—are uniquely equipped to build. If Europe can nurture this homegrown competitor while leveraging its regulatory environment, it may yet carve out a significant stake in the coming humanoid revolution.