A Parisian company named UMA, founded by Rémi Cadène, a Frenchman who worked at Tesla and Hugging Face, unveiled its humanoid robot Northstar and its in-house architecture called Real-Time Learning at the Machina Summit held at Station F on July 7th. The robot learns a gesture by watching a human demonstration and repeats it without any scripted trajectories or manual programming. Cadène, who worked on Autopilot AI and built the first neural networks for Optimus at Tesla between 2021 and 2024, then built LeRobot at Hugging Face. The team includes Pierre Sermanet, a researcher from Google Brain and DeepMind, and Robert Knight, founder of The Robot Studio. The company is backed by a seed round of around $40 million, with investors including Yann LeCun, Thomas Wolf from Hugging Face, Xavier Niel, and Nico Rosberg. The Northstar robot has a deliberately lighter build than a human, wheels instead of legs, a neutral visor, and two planned variants: an industrial dual-arm version for warehouses and a compact one for hospitals and labs. The first prototype was assembled in Paris in 9 months. UMA’s strategy focuses on Europe first, betting that a robot that learns fast is worth more than a robot that does backflips.