Bleu Robotics, a Paris-based startup building AI software for industrial humanoid robots, has raised €5 million in pre-seed funding co-led by Contrarian Ventures and OVNI Capital. Kvanted, WEPA Ventures, Motier Ventures and Meta board member Charlie Songhurst also joined the round, announced October 29.
The company, founded in December 2025, is not building its own robot. Instead, it sells a learning layer it says can teach any humanoid platform a new factory task from as few as ten demonstrations by a human operator, using handheld grippers and no coding. Bleu calls the approach abductive learning: rather than training on vast datasets, the system infers an operator's intent, goals and constraints from a small number of examples and generalizes from there. The company says the method holds production-line accuracy above 99% and can take a robot from zero to a deployed new task in under an hour.
Bleu Robotics was founded by Benoît Berkoukchi, a former Renault assembly-line manager, alongside Jean-Baptiste Mouret and Serena Ivaldi, both research directors at French science agency INRIA. The startup is already working with two large industrial companies it has not named and plans to use the new capital to expand its research and robotics teams and bring its first commercial learning models to several industries.
Why Few-Shot Learning Matters for Humanoid Robots
Most humanoid robot makers, including Figure, Agility Robotics, Unitree and AgiBot, are racing to build general-purpose foundation models trained on enormous volumes of demonstration and simulation data. That approach is expensive and slow to adapt when a factory changes a task or a product line.
Bleu is betting that manufacturers will pay less for a model that trains fast on-site than for a model that trains slowly in a lab. The pitch echoes microagi, a Munich startup that raised $55 million earlier this year to record factory workers performing tasks and convert that footage into training data for existing robot models. Both companies treat the robot hardware as a commodity and compete on how quickly software can turn a line worker's actions into a working robot skill.
"Humanoid robots in manufacturing are the next automotive sector, with potential to help Europe and the US reshore production," said Berkoukchi. Thomas Poitrineau of Contrarian Ventures said the founders' ambitions extend well beyond building the best model in a lab, pointing to Bleu's plan to make its software run across competing hardware rather than lock customers into one robot brand.
A Crowded Field for Robot Training Data
The deal lands as humanoid robot production accelerates, particularly in China, where domestic manufacturers built more than 400 models and shipped over 40,000 units in the first half of 2026. That volume creates demand for software that can retask a fleet quickly without retraining from scratch for every new assembly line or SKU change, which is the gap Bleu says its abductive-learning method fills.
Bleu's €5 million is modest next to the hundreds of millions raised by foundation-model rivals, but pre-seed rounds in industrial humanoid software have been getting larger as investors look past the robots themselves toward the software deciding what they do. The company says its next milestone is converting its two pilot deployments into paying production contracts while expanding the number of robot platforms its software supports.
Source: Robotics & Automation News