Most industrial robots are built for force and speed, not for touch. They excel at welding car frames or lifting pallets, but ask one to pick up an egg or adjust a bandage and the results are predictable. This is the gap a Swiss startup, Embodied AI, is trying to close with a hardware-first approach. The company is developing soft robotic manipulators, essentially robot hands made from compliant materials, paired with AI models designed to translate simple human instructions into safe, physical actions [Look AI Ventures, retrieved 2024].
It is a bet on a different kind of robot intelligence, one where safety and intuitive interaction are the primary design constraints. The company, a spin-off from ETH Zurich founded in 2023, has secured over CHF 1 million in non-dilutive funding, including a CHF 150,000 grant from the Swiss accelerator Venture Kick [embodiedai.ch, retrieved 2024] [venturekick.ch, retrieved 2026].
The technical wedge: compliance by design
The core of Embodied AI's proposition is a shift from rigid to soft manipulation. Traditional robotic grippers use precise, stiff mechanisms that require complex force-feedback systems to avoid crushing objects. Embodied AI's approach starts with compliant materials in the end-effector itself, which can absorb shocks and conform to shapes passively. This physical compliance is then augmented by AI models for vision-language-action (VLA) and motion planning [embodiedai.ch, retrieved 2024].
A team built in the lab
The founding team reflects the project's deep academic roots. CEO Francesco Stella holds a PhD in Robotics and AI from EPFL, with research stays at MIT and TU Delft [aiforgood.itu.int, retrieved 2026]. He previously founded and led Helix Robotics, which rebranded to become Embodied AI [Crunchbase, retrieved 2026]. He is joined by co-founders including CTO Kai Junge and COO Max Polzin [dealroom.co, retrieved 2026] [tracxn.com, retrieved 2026].
The crowded field of embodied intelligence
| Company | Primary Focus | Key Differentiator |
|---|---|---|
| Sereact | AI-powered pick-and-place software | Software-only solution for existing robot arms |
| Apptronik | Humanoid robotics (e.g., Apollo) | Full-body humanoid platform for logistics |
| Wayve | Embodied AI for autonomous vehicles | Geographic intelligence for self-driving cars |
| Embodied AI | Soft robotic manipulators & AI | Proprietary compliant hardware + AI stack |
The scale-up questions
From a technical standpoint, the challenges ahead are clear. Soft robotics introduces material science hurdles: durability, consistency, and cost. The AI models face the "sim-to-real" transfer problem common to all robotics. Financially, the path is steep. The next milestone will be closing a substantive pre-seed or seed round to fund the jump from research prototypes to pilot-ready units.