Bleu Robotics
Track: raisedWebsite ↗𝕏 @BleuRobotics01LinkedIn ↗
potential 76 · scoop 82
Bleu Robotics develops software that trains humanoid robots on factory tasks from under ten operator demonstrations. The Inria spinout closed a €5 million pre-seed round led by Contrarian Ventures and OVNI Capital.
Scores run 0 to 100. Potential: how likely the firm is to become a real winner, weighing all six dimension scores, whether or not anyone has heard of it. Scoop: how unknown the firm still is to a well-read tech investor. High means the outlets that investor follows have not covered it yet, however good or old the firm is. How our top potential scores did.
A former Renault assembly-line manager joined two Inria researchers to raise €5 million for software that teaches robots factory tasks from ten demos.
Snapshot
- What they build
- AI models for industrial humanoids
How we found them
- Surfaced in issue
- Nov 2, 2026
- HQ
- France
- Discovery track
- raised
- Discovery context
- €5m for AI models for industrial humanoids
Dimension scores
- Team 85 out of 100
- Product 70 out of 100
- Traction 55 out of 100
- Market 80 out of 100
- Moat 65 out of 100
- Capital 75 out of 100
Analysis
Jean-Baptiste Mouret and Serena Ivaldi left their Inria Nancy posts to co-found the company with Benoît Berkoukchi. Mouret created the MAP-Elites algorithm and led the 2015 Nature cover paper on resilient robots. Ivaldi headed the HuCeBot humanoid team. Berkoukchi managed Renault production lines. Roughly half the first eight hires came from the same Nancy research group.
The software infers task goals and constraints from a few demonstrations rather than scaling imitation data. It leaves Cartesian controllers outside the learned model. Training finishes in minutes on a laptop. Lab tests succeeded on a Unitree G1 and a Franka arm. The €5 million round closed in late October 2026.
Humanoid hardware output exceeded 20,000 units in 2025, but only about 10 percent reached active use. EU and US reshoring policies plus labor shortages now favor quick redeployment tools for high-mix lines. The team positions the product as a narrow industrial wedge rather than another generalist model.
What they do
Bleu Robotics develops abductive-learning AI software and training devices that let operators teach any ROS-compatible industrial humanoid (e.g., Unitree G1) a new factory task from ~10 demonstrations, targeting full workstation deployment or redeployment in under an hour with explainability and >99% accuracy for production lines.
Why now
Humanoid hardware scaled rapidly in 2025 (>20k units produced, but only ~10% reached real-world use), exposing the deployment-time bottleneck that traditional integrators or data-hungry models cannot solve economically in factories. Recognition that factory-specific data will remain scarce favors few-shot, intent-inferring methods. EU/US reshoring and labor-shortage pressures create pull for rapid-redeployment automation in high-mix lines.
Founding story
Incorporated December 22, 2025 in Paris (SIREN 995292422) as a spin-out of Inria’s HuCeBot humanoid-robotics team in Nancy. The three founders combined Inria’s 15 to 20+ years of robot-learning/humanoid research with Berkoukchi’s Renault manufacturing operations experience. The October 2026 OVNI Capital podcast frames the origin around the recognition that factory deployment, not hardware or generalist models, is the bottleneck, and that scarce real-world factory data favors data-efficient, explainable methods over internet-scale approaches.
Competitive position
Occupies a distinct lane, hardware-agnostic software specifically optimized for industrial task teaching/retraining with minimal data and high explainability, positioning as the “last-mile” enabler rather than another robot body or generalist brain. Closest startup peers: other few-shot/imitation or deployment-focused robotics AI firms.
Defensibility
Strongest elements are the 15 to 20+ year Inria research pedigree (MAP-Elites, resilient learning, teleoperation, whole-body control) plus the rare researcher + manufacturing-operator founder combination. Potential switching costs once integrated into factory workflows; explainability/safety positioning for industrial certification; early data from real deployments could create a task-specific flywheel. Overall moat is pedigree + vertical insight rather than insurmountable technical or network effects at this stage.
Network density
Early team of ~8 includes multiple individuals with Inria/HuCeBot ties (e.g., Waldez Gomes, Roman Klokov, Alexandros Paraschos). Roughly 3 to 5 of the initial ~8 to 10 named individuals trace to the Inria Nancy/Loria/HuCeBot ecosystem (~40 to 60% early technical hires from the same prior research group).
Early team signals
Team of 8 with active hiring (5 roles listed Oct 2026). High Inria Nancy/Loria/HuCeBot ecosystem density (~40 to 60% early technical hires).
Traction
- €5m pre-seed from tiered European VCs + high-profile angel (Songhurst)
- Live website with Bleu_0.1 model (March 2026) and technical claims
- Lab validation on Unitree G1 humanoid and Franka arm
- Team of 8 with active hiring (5 roles listed Oct 2026)
- Partnerships planned with robot distributors/integrators
- Coverage across robotics press (The Engineer, EU-Startups, Robotics & Automation News, Humanoids Daily, Contrarian substack)
Opportunities
- EU/US reshoring and labor-shortage tailwinds accelerating demand for flexible automation
- Continued commoditization of humanoid platforms (Unitree et al.) lowering hardware barriers
- Apparent limits of pure end-to-end/generalist models in data-scarce factory settings
- Potential distribution via robot makers/integrators with recurring subscription revenue
- Data flywheel from real deployments refining the abductive models
Risks / red flags
- Performance claims (98 to 99% success on ManiSkill3 StackCube with <10 demos; <2 min training; 100× data efficiency; sustained >99% production accuracy; <1-hour full deployment) rest on lab/sim benchmarks and early tests, not yet triangulated with named, sustained factory production data
- Pre-commercial stage; customers described only as “two unnamed major manufacturers in automotive and consumer goods”
- Early founding (Dec 2025), pre-seed funding, and heavy reliance on two Inria researchers plus one ops founder create key-person and execution concentration risk
- Unnamed logos and “first deployments under way” language
- Crowded humanoid software lane with well-funded players
- Regulatory exposure (industrial safety standards, explainability mandates) not yet visibly addressed beyond claims
Hidden bets
- Abductive learning (inferring task goal, constraints, and simplest explanatory process from few demonstrations) rather than end-to-end diffusion or large-scale imitation policies
- Training/inference runs on modest hardware (laptop minutes for Bleu_0.1)
- Hybrid architecture keeps traditional Cartesian/whole-body controllers outside the learned model for reliability
- Handheld gripper training devices for on-site, no-code demos
- Explicit explainability and safety focus for industrial certification
- Narrow, deployable wedge for high-mix/low-volume manufacturing on commodity platforms (Unitree G1, Franka, ROS)
Key relationships
- Contrarian Ventures (co-lead)
- OVNI Capital (co-lead)
- Kvanted
- WEPA Ventures
- Motier Ventures
- Charlie Songhurst (Meta board member/tech investor)
- Lukas M. Ziegler
Founders
Citations
- https://bleu-robotics.com/index
- https://x.com/BleuRobotics01
- https://fr.linkedin.com/company/bleu-robotics
- https://www.frenchtechjournal.com/bleu-robotics-5m-humanoid-robot-software/
- https://www.eu-startups.com/2026/10/french-startup-bleu-robotics-secures-e5-million-to-teach-humanoid-robots-new-tasks-with-fewer-demonstrations/
- https://www.societe.com/societe/bleu-robotics-995292422.html
- https://podcast.fr/episode/ovnis-ep-129-benoit-berkoukchi-lusine-du-futur-aura-t-elle-encore-besoin-dhumains/
- https://members.loria.fr/JBMouret/resilient.html
- https://scholar.google.com/citations?user=lp8V_UYAAAAJ&hl=en
- https://github.com/jbmouret
- https://members.loria.fr/SIvaldi/
- https://members.loria.fr/SIvaldi/cv/
- https://contrarianventures.substack.com/p/a-research-lab-on-the-factory-floor
- https://rocketreach.co/bleu-robotics-profile_b6b83bb7c826ac37
- https://team.inria.fr/hucebot/team/serena-ivaldi/
- https://www.theengineer.co.uk/content/news/bleu-robotics-secures-5m-for-humanoid-robot-ai
- https://www.eu.vc/p/bleu-robotics-raises-5m-to-speed-factory-robot-training
- https://humanoid.guide/bleu-robotics-raises-e5-million-for-humanoid-task-training/
- https://theaiinsider.tech/2026/10/30/bleu-robotics-raises-e5m-in-funding-to-speed-humanoid-robot-deployment-in-factories/
- https://www.todaysstartupnews.com/startups/bleu-robotics-raises-5-million-pre-seed-contrarian-ovni-humanoid-robot-learning
- https://www.humanoidsdaily.com/news/bleu-robotics-5m-industrial-humanoid-training