INVESTMENT NOTE
Embodied AI
Robots that can safely learn on the job.
THE PROBLEM
Most industrial robots work well only in tightly controlled settings. They are rigid, difficult to reprogram and poorly suited to delicate objects or close interaction with people. The next wave of automation needs bodies that are safer around humans and intelligence that can adapt to real work.
THE APPROACH
Embodied AI pairs compliant robot hardware with AI that translates human demonstrations and language into physical actions. The company began at EPFL as Helix Robotics, developing a soft manipulator for food handling, laboratories and other shared environments. Its current strategy uses teleoperated humanoids to do useful work while collecting real-world data that can improve manipulation skills over time—a practical path from human supervision to greater autonomy.
WHY THIS TEAM
CEO Francesco Stella earned a PhD in robotics and AI at EPFL and spent time at MIT. CTO Kai Junge’s EPFL doctorate focused on dexterous manipulation; he co-authored peer-reviewed work on compliant anthropomorphic hands and intuitive teleoperation. COO Max Polzin was a technical lead at Seervision, the Swiss AI-camera company acquired by Q-SYS in 2023. The underlying research also grew from work with EPFL professor Josie Hughes and TU Delft professor Cosimo Della Santina.
MOMENTUM
The team received CHF 150,000 from Venture Kick in January 2025, placed in the top three of the 2025 >>venture>> Industrials & Engineering awards, published results in Nature Portfolio journals and entered ESA BIC Switzerland for 2026–2028. In 2026, Google DeepMind selected Embodied AI as one of 15 companies in its inaugural EMEA Robotics Accelerator. Public deployment and financing details remain limited.