Bleu Robotics raises €5M pre-seed to teach humanoids factory tasks

Paris-based Bleu Robotics has raised a €5M pre-seed round, co-led by Contrarian Ventures and OVNI Capital, to build software that teaches industrial humanoid robots a new task from about ten human demonstrations.

Five funds and one angel behind the €5M

Bleu Robotics announced the round on October 29, 2026, as reported by Vestbee. Contrarian Ventures and OVNI Capital co-led it. Kvanted, WEPA Ventures and Motier Ventures joined, along with technology investor Charlie Songhurst.

The company says the money will grow its research and robotics teams and take its first industrial robot-learning models into deployment. It has not disclosed a valuation.

Bleu Robotics was founded in 2025 and spun out of Inria, the French national research institute for digital science, according to CFNews. Benoît Berkoukchi, who previously managed assembly-line teams at Renault, founded it with two Inria robotics researchers, Jean-Baptiste Mouret and Serena Ivaldi.

Ten demonstrations with a handheld gripper

An operator performs the task about ten times while holding a training gripper. From those recordings, the software works out the goal of the task and the constraints around it. Vestbee describes the method as abductive learning. The company says this cuts the time to deploy a humanoid on a new task from months to under an hour, with no large training dataset and no code.

Anyone who has sat through a factory automation project will recognize the months in that sentence.

The software is built to run on more than one robot. Humanoids Daily reported demonstrations on a Unitree G1 humanoid and a Franka Research 3 arm, and noted two limits the team states openly. The system handles rigid objects only for now, so textiles are out of scope. When it meets a situation it does not recognize, it stops and asks a human for help.

Bleu Robotics says it is working with two major industrial companies, which it has not named. First deployments are planned with automotive and consumer goods manufacturers, and the company is building partnerships with robot distributors. Sifted reported that the business model is a license sold to robot makers and integrators, billed as a monthly subscription per robot.

A $15bn humanoid market by 2030, on one forecast

MarketsandMarkets sized the global humanoid robot market at $2.92bn in 2025 and forecasts $15.26bn by 2030, a 39.2% CAGR (report published April 2025). That figure counts robot sales. A per-robot software subscription sits on top of the installed base, so Bleu’s addressable market grows only as fast as humanoids leave the pilot stage and stay on the line.

Two rounds on our own records give the amount scale. Enact Intelligence raised a $30M pre-seed for robots that feel force, led by redalpine and announced on October 22. Five days later, Machenta raised a $40M seed to teach robots from one demonstration, co-led by Creandum and Union Square Ventures. At €5M, Bleu is the smallest of the three by a wide margin.

Three robot-learning rounds in eight days

Three European teams have now raised money in eight days to solve the same problem, which is how a robot picks up a new task without months of engineering. I read that as a sign that investors here have stopped waiting for the hardware question to be settled.

The humanoid Bleu demonstrated on, the G1, is made by Unitree in China. The good news is that the software layer travels across robot bodies, and the automotive and consumer goods plants where Bleu plans its first deployments are a European strength we already have. A Renault line manager and two Inria researchers in Paris are a reasonable founding team for a product that has to work on a real shift, with real tolerances.

The investor mix adds a second signal. Contrarian Ventures is a climate seed fund on our records, writing €1M to €4M checks across Europe and Israel. A climate fund co-leading a humanoid software round shows how wide the industrial thesis has become at seed stage.

Imagine if the first humanoids that hold a full shift on a European line run on software trained in Paris, on hardware built elsewhere. That would be a very European outcome, assembled from parts we did not all make.

What the next twelve months should show

I will be watching for three things. The two industrial partners going public by name would be the first. A move from rigid parts to flexible materials would be the second. The third is whether a per-robot monthly price holds when a plant runs dozens of units. Our fundraising data will show how many more teams chase the same problem before the end of the year.

Let’s see those ten demonstrations hold up on a real factory line!