ProYarn

Deals · Physical AI / Robotics

Bleu Robotics raises €5M pre-seed to put industrial humanoids to work in one hour

A Paris-based Physical AI startup that lets industrial humanoid robots learn new tasks from as few as ten operator demonstrations using a proprietary abductive learning approach, with a stated deployment cycle of under one hour.

ProYarn Desk · Read this in French
in𝕏✆
“Humanoid robots in manufacturing are the next automotive sector, with the potential to help Europe and the US reshore production. But most robotics companies are trying to train humanoids with millions of data points and build fully general-purpose models. That approach isn't suited to manufacturing. There isn't enough data on factory work, and there won't be internet-scale data, the kind LLMs were built on, for robots for a long time. Our methodology means we can do more with less.”
Benoît Berkoukchi — Co-founder, Bleu Robotics

Bleu Robotics (Paris) has raised €5 million in a pre-seed round co-led by Contrarian Ventures and OVNI Capital, with Kvanted, WEPA Ventures, Motier Ventures, and Charlie Songhurst also participating. Founded in 2025 by Benoît Berkoukchi, Jean-Baptiste Mouret, and Serena Ivaldi — two INRIA researchers and an industrial engineer — the company builds AI software that teaches industrial humanoid robots new tasks from as few as ten operator demonstrations and, it claims, deploys them in under an hour.

Why today's humanoids don't reach factories

The industrial humanoid market has an odd property: the hardware is arriving faster than it can be used. More than 20,000 humanoids were produced globally in 2025. Only about 10% reached real-world applications — not because robots aren't capable enough, but because getting a specific robot to perform a specific factory task remains a months-long, heavily data-intensive process.

The dominant software model is end-to-end learning: train a large neural network on enormous volumes of demonstration data until the robot can generalise. It works at internet scale — for language and images, where training data is abundant. Factory floors are not internet-scale environments. High-mix, low-volume production changes constantly; the same task may look different depending on part tolerances, lighting, tooling variation, and which operator is on shift. There is no LLM-scale dataset for that, and Bleu Robotics argues there will not be one.

The company's answer is abductive learning: rather than inferring all behaviour from raw inputs in a single end-to-end model, the system captures the observable reasoning structure of what an expert operator does — the sequence of checks, the decision points, the corrections — and encodes it into a model the robot can execute. The company reports 99%+ accuracy at the industrial safety threshold and a learning cycle of as few as ten demonstrations per task.

Its software is hardware-agnostic, running on any humanoid platform and including the AI model and the training devices used to teach the robot on the factory floor.

What the €5M funds

The pre-seed capital goes to expanding the research and robotics engineering team and scaling the first industrial deployments — currently underway with automotive manufacturers and with WEPA Group, the European hygiene-paper producer whose investment arm, WEPA Ventures, appears in the syndicate as both investor and customer.

"Humanoid robots in manufacturing are the next automotive sector," said Benoît Berkoukchi, co-founder of Bleu Robotics. "Our methodology means we can do more with less."

Contrarian Ventures has led rounds across European deep-tech; its presence here alongside OVNI Capital — which has backed three ProYarn-tracked French deals including Epyr and MokN — gives the cap table both cross-border pattern-matching and French-market depth. Motier Ventures, which also co-invested in Mallow, adds a further French-ecosystem thread. That three investors from distinct parts of the European venture network are converging on the same French robotics pre-seed is the round's clearest signal that the industrial humanoid software thesis is being taken seriously before any of the hardware manufacturers have built their own data infrastructure internally.

The ten-demonstration claim, if it holds across a full task library at production scale, removes the data moat that has kept industrial humanoid deployment slow. Whether it does is the question the next eighteen months will answer.

Sources

  1. 01French startup Bleu Robotics secures €5 million to teach humanoid robots new tasks with fewer demonstrations — EU-Startups
  2. 02Bleu Robotics Raises €5M in Pre-Seed Funding — FinSMEs
  3. 03The French robotics startup promising to deploy industrial humanoids after just 20 minutes of training — Sifted

Threaded to this story

Every European round, in your inbox by 8am.

The day's seed and Series A rounds across France and Europe — threaded, sourced, and read in two minutes. Free.

Double opt-in. We'll send one confirmation email. Unsubscribe anytime.