Swiss startup Embodied AI launches AIR, a humanoid robot that learns from factory workers
Hello Air!
Most humanoid robot companies start with a demo. Embodied AI started with a paying customer.
The Lausanne-based startup, an EPFL spin-out founded in 2023 by PhD roboticists from EPFL and MIT, unveiled its first full humanoid robot today at Humanoid Europe 2026. AIR is a wheeled upper-body humanoid designed for factory floors:
- tendon-driven arms
- all heavy components centralized in the torso
- a learning architecture that makes it productive via teleoperation from its first shift
The company simultaneously announced its first institutional funding round, led by Faber VC with participation from Techshop Capital, Look AI Ventures, Kickfund, Plug and Play, and others.
The amount was not disclosed.
Why wheels and why tendons
Embodied AI is making a few explicit choices that push against the prevailing industry trend toward bipedal, general-purpose humanoids.
- Wheels instead of legs: factory floors are flat. Legged locomotion adds mechanical complexity and failure modes without adding value in a production environment. Wheeled bases are more stable, more reliable, and far easier to certify under industrial safety standards.
- Tendon-driven arms: by routing cables from motors housed in the torso to joints in the arms, the arms themselves carry almost no moving mass. The result is arms that give way on contact through intrinsic mechanical compliance rather than software-enforced torque limits. If a worker steps into the robot's path, the arm absorbs the impact physically rather than relying on sensors and algorithms to react in time.
It's also worth noting: wheels are easier for a small team just starting out.
AIR is designed to meet ISO/TS 15066:2016, the international standard for collaborative robot workspace safety. CE marking is planned for 2027.
The case for physical safety
Embodied AI says safety is a core consideration:
"Hallucination of AI models is acceptable in the digital space because humans review, reject, or correct the output before acting on it. In robotics, however, incorrect actions have immediate and irreversible physical consequences," says CEO Francesco Stella. "When AI is used to take physical actions, errors can cause severe damage or endanger human lives. Safety and reliability are non-negotiable requirements of physical AI to enter the real world at scale."
This is a direct rejection of the approach that trains general-purpose AI models to exhaustion in simulation and then deploys them hoping for the best.
Embodied AI's answer is to design the hardware so that failures are physically constrained, then let the AI catch up over time.
Productive on day one, autonomous within a month
The learning architecture follows from the same philosophy.
AIR is teleoperated by a human operator from the first shift, generating immediate productive output for the factory while every run is recorded. The company's proprietary AI stack learns from those accumulated demonstrations until AIR can perform the task autonomously, which Embodied AI says typically takes about a month.
After the robot achieves autonomy, the human operator stays available to handle edge cases the robot has not yet encountered. What one robot learns can transfer to other robots in the fleet.
The company calls this architecture the "Embodied AI Flywheel": rather than training a single large general-purpose model, it builds small task-specific AI models trained efficiently during production while the robot maintains uptime.
Each deployment generates data for the next one.
Already in factories
AIR is a formal product launch, but Embodied AI is not starting from scratch commercially. The company's earlier two-arm industrial cell is already in paid deployments with European manufacturers on tasks including consumer electronics assembly, kitting, flexible cable handling, and automotive assembly.
The cable-handling result is worth highlighting.
When a leading consumer electronics manufacturer tested 5 funded robotics startups on a cable-handling task, Embodied AI was the only one to meet all performance targets, achieving 98% task reliability, against a 90% goal. Cable handling is notoriously difficult for robots because cables are deformable and unpredictable.
The company says it has signed contracts with major OEMs globally.
The European humanoid bet
Embodied AI says it's filling a gap.
The humanoid robot industry is dominated by US and Chinese companies: Figure, Physical Intelligence, Boston Dynamics, Tesla, Agility, Unitree, AgiBot, Fourier. European industrial automation has strong players in collaborative arms (ABB, KUKA, Fanuc) but fewer humanoid robot startups (although it has strong players in Neura Robotics, Agile Robots, and Humanoid AI, among others).
"By vertically integrating proprietary soft-robotic hardware, teleoperation and autonomy-ready software, Embodied AI is creating a fundamentally new robotics platform," says Faber VC partner Sofia Santos.
And it looks pretty funky, too.
Sources
Written by
I tell stories with data and build startups. I've raised angel and VC funds, led software companies and startups, and chronicled the rise of the mobile economy. Currently I write for Forbes, host several podcasts, and am starting up HumanoidMegaHub. I was TUNE's mobile economist and Singular's VP of Insights, built the research team at VentureBeat, and managed teams creating software for partners like Intel and Disney. In past lives I built websites, e-learning platforms, and mobile apps, and, a long time ago, was named to Folio's top 100 of the media industry's "most innovative entrepreneurs and market shaker-uppers.”