Physicl has raised $10 million in seed funding to help robots and world models learn from the rare situations that break them in the real world. Inovia Capital led the round, with US Venture Partners taking part.
The Paris and San Francisco based startup announced the round on 6 November 2026, alongside the launch of Studio, its first product. Studio lets researchers recreate the exact moments their models get wrong and generate thousands of variations of each one.
The gap Studio is built to close
Every leap in artificial intelligence has followed a leap in data, and physical AI has no comparable corpus to learn from. Language models trained on the text of the internet, while image models trained on billions of captioned photos, but the physical world does not come as a dataset.
The gap shows up at the end of a project, when a robot that succeeds nine times out of ten still cannot be deployed next to people.
The failures come from rare moments, like a cup half hidden by a bag or a shadow crossing a counter at sunset, and those moments are almost impossible to capture in real life.
Even when a team does catch one, it cannot replay the moment or vary it. Physicl builds the worlds instead, from geometry and physics that are known, so every label is exact and every condition can be changed on purpose.
What Studio does
Studio starts with a situation a model gets wrong and generates it thousands of ways, with exact ground truth on every frame. Depth, surface normals, segmentation, boxes and camera pose come from the scene itself, not from estimates made after the picture is taken.
Researchers can change one condition at a time, moving the light, the clutter, the camera or the objects alone or together. The library behind it holds 10,000 environments and more than 100,000 assets built from real geometry with known materials.
The product is in beta and free during that period, with access opening to research teams in waves. Labs, universities and companies building vision, world model and robotics systems can request access through the company site.
A founder who has built this before
Physicl was founded by Alex de Vigan, who spent a decade building nfinite, a company that creates 3D product visuals for retailers and brands. He founded nfinite in 2017, and it raised a $15 million Series A in February 2022 led by US Venture Partners.
Four months later, nfinite raised a $100 million Series B led by Insight Partners, with US Venture Partners taking part again. The same two investors now back Physicl, which gives the new venture a funding base that already knows the team and its technology.
De Vigan has said the arrival of transformer based AI pushed nfinite toward generative imagery. At the same time, robotics labs, world model labs and vision teams started calling with a different problem.
They did not need prettier pictures. They needed exact data on the situations their models failed in, and they needed thousands of examples. That request became Physicl.
The company works with AI labs including Meta, World Labs, Epic Games and SpAItial, according to its announcement. Its advisers include Thomas Wolf of Hugging Face, Christoph Lassner of World Labs and Matthias Niessner of the Technical University of Munich.
We believe physical AI represents one of the most significant emerging opportunities in technology, but the industry still faces fundamental challenges in translating model capabilities into real-world performance. Physicl is tackling a critical piece of that puzzle: scalable, high-quality physical data grounded in real-world physics.
Steve Woods, partner at Inovia Capital
Early proof from a first customer
SpAItial, which builds world models that turn real rooms into explorable 3D spaces, used Physicl data to train its models. It more than halved its reconstruction error on real rooms, in an evaluation by the team of Matthias Niessner at the Technical University of Munich.
The result matters because SpAItial released Echo-3, its latest world model, last week, and fidelity is the product. Physicl wants to be more than a data provider, and it describes Studio as tooling for ongoing synthetic data generation rather than a one off dataset.
You’re running Claude, and it checks, ‘Oh, I need more data of this in my training,’ and then it auto generates that data, depending on the training state. That, to me, is super exciting, and this is exactly what Physicl is offering.
Matthias Niessner, chief executive of SpAItial and professor of computer science at the Technical University of Munich
Why the timing matters
Physicl enters a crowded field, and de Vigan says so himself. He describes the space as very young and very blurry at times, with many companies collecting, labelling or generating data in different forms while customers work out what makes robots live up to their promise.
The wider field is moving fast, and the money behind it is large. World Labs, the spatial intelligence startup founded by Fei-Fei Li, raised $230 million in September 2024 and a further $1 billion in February 2026, before AMD agreed to acquire it for $8.2 billion in September 2026.
Robotics funding is moving in the same direction. KeeVurds covered a $51 million Series A for autonomous hauler builder Humble Robotics on 6 November.
The seed money funds the team in Paris and the San Francisco Bay Area and the platform behind Studio. The next test is whether research teams adopt a tool that generates training data on demand, instead of collecting it by hand.
