€5m · Pre-Seed · Robotics · Paris, France
Paris-based Bleu Robotics, which develops software for teaching industrial humanoid robots new factory tasks, has raised €5m in pre-seed funding. Contrarian Ventures and OVNI Capital co-led the round, with Kvanted, WEPA Ventures, Motier Ventures and technology investor Charlie Songhurst participating. The company says the capital will help turn its first robot-learning model into a product that can be deployed in factories.
Teaching the task instead of rebuilding the cell
Bleu is targeting a costly part of factory automation: adapting a robot when the product, workstation or sequence changes. Conventional cells can require specialised grippers, vision systems and engineering for each application. Bleu’s approach asks an operator to demonstrate the job with handheld training grippers, then uses what it calls abductive learning to infer the simplest process that explains those demonstrations. Its technical description of Bleu_0.1 says the model can learn from fewer than ten demonstrations, train in under two minutes on a laptop and run across robots that use the ROS standard.
The commercial mechanism is a reusable software layer rather than another humanoid body. Bleu plans to distribute its model through robot makers and systems integrators, with manufacturers paying a monthly subscription for the software, new task training and model updates. If task knowledge and integrations transfer across hardware and sites, the company can reduce the engineering attached to each deployment. If every factory still needs a bespoke project, the software economics will look much closer to traditional systems integration.
Factory evidence is the next milestone
The product is still pre-commercial. Bleu says it is working with an automotive manufacturer and a consumer-goods company on initial factory deployments, but it has not named those customers or claimed production-scale use. The company told The French Tech Journal that its near-term goal is a limited but useful model tested in a closed loop on real factory floors. The company’s published limitations are equally important: Bleu_0.1 focuses on rigid objects, stops when a trajectory falls outside its training distribution and still needs work on precise manipulation, teaching interfaces and movement between workstations.
That makes the pre-seed round a deployment test as much as a research bet. Bleu combines two Inria robotics researchers with chief executive Benoît Berkoukchi’s manufacturing experience at Renault, and reports strong results on a simulated benchmark. The evidence that matters next is whether the team can reproduce that reliability across longer, changing tasks in customer plants while keeping operator intervention and integration work low enough for a subscription model.
Explore this theme
Physical AI market map — Synaptic, adapted by EUVC · 9 April 2026. The map explains why scarce real-world interaction data, deployment economics and the gap between simulation and factories are central tests for companies building intelligence for machines.
