Bleu Robotics Raises €5M to Teach Industrial Humanoids New Tasks in Under an Hour

Paris-based Bleu Robotics has raised €5 million in pre-seed funding to develop software that it says can teach industrial humanoids new tasks in less than an hour. Contrarian Ventures and OVNI Capital led the round, with participation from Kvanted, WEPA Ventures, Motier Ventures and technology investor Charlie Songhurst, according to Sifted.

Key Takeaways

Show all

• Bleu has raised €5 million to turn its robot-learning research into a product for industrial customers.
• Its claimed workflow involves about 20 minutes gathering human demonstrations, followed by another 20 minutes before the robot performs the task.
• The company is building software for different robot platforms, with the funded product still being developed for industrial deployment.

The startup targets workstations where products and processes change frequently, making dedicated automation difficult to justify. Its ambition is to let factory operators teach robots new jobs without commissioning a separate engineering project for every changeover.

“We raised to build a product on top of this technology,” co-founder Benoît Berkoukchi told Sifted. Berkoukchi previously managed automotive assembly-line teams at Renault; fellow co-founders Jean-Baptiste Mouret and Serena Ivaldi bring robotics research experience from Inria.

What the 20-minute claim means

In the workflow Berkoukchi described, an operator demonstrates a task using handheld training grippers. Gathering that data takes around 20 minutes, and the robot is working at the workstation another 20 minutes later, he said. Training and inference run onboard rather than in the cloud.

An earlier engraving-workstation demonstration shows an operator using a UMI-style handheld gripper, followed by a dual-arm mobile robot working at the same station. This fits the broader approach explored in our UMI feature: collecting manipulation demonstrations with a human-held device.

That makes the pitch a roughly 40-minute demonstration-to-execution process.

Berkoukchi told Sifted that around ten demonstrations can be enough to train a robot for an industrial workstation, without pre-training on that task.

Learning the task without learning everything from scratch

Bleu’s technical introduction to Bleu_0.1 describes “abductive learning”: finding a simpler representation of the demonstrated behavior so that fewer examples are needed.

The system combines task learning with existing capabilities for tracking objects in three dimensions and controlling robot motion. Its framework explicitly includes pre-trained perception models, so the claim about avoiding pre-training does not mean every part of the system starts without prior knowledge.

The company presents drilling-workstation demonstrations using a Unitree G1 and a Franka Research 3 arm, illustrating its aim to support different hardware. It also identifies limits: rigid objects are the main focus, textiles remain outside its current scope, and it says the system stops and requests help when faced with an unrecognized situation. Autonomous movement between workstations is listed as further development work.

From research to factory deployments

Bleu has initial industrial partners lined up, Berkoukchi told Sifted, with plans to develop the product through regular access to their factories. The earlier engraving experiment shows work in an industrial setting; the next step is turning that research into a product for routine factory use.

Longer term, Bleu plans to sell its software to robot manufacturers, with systems integrators handling customer installations. The proposed business model is a monthly subscription per robot covering training for new tasks and access to updated models.

The commercial test will be whether short training sessions translate into reliable work across changing parts, workstations and production conditions. The funding gives Bleu resources to pursue that transition; the deployment-time claim remains a company claim to prove in factory operation.