$10m · Seed · Software · Paris, France
Paris-based Physicl has raised $10m in seed funding to expand the team and platform behind Studio, its synthetic training-data product for physical AI. Inovia Capital led the round, with US Venture Partners participating.
Studio lets robotics, computer-vision and world-model teams describe a scene their model handles poorly, vary conditions such as lighting, clutter, object placement or camera position, and render labelled training examples. Physicl says each frame can include depth, surface normals, segmentation, bounding boxes and camera pose derived from the underlying 3D scene.
Controlled variation is the product
The commercial proposition is more specific than supplying a larger generic dataset. A model team can hold most of a scene constant, change one troublesome condition and compare performance across repeatable runs. That can make it easier to diagnose an edge-case failure and generate targeted examples without waiting to capture the same rare situation in the real world.
Physicl says Studio starts with more than 10,000 environments and 100,000 objects. The library gives the company a base for generating many controlled variants, but coverage and realism will still need to match each customer's deployment environment. Exact labels for a rendered scene do not by themselves prove that a robot or vision model will generalise to the physical world.
The company reports that SpAItial more than halved reconstruction error on real rooms after training with Physicl data, in an evaluation by Matthias Niessner's team at the Technical University of Munich. That is an encouraging company-reported case study rather than an independent benchmark, and Physicl has not disclosed revenue or post-beta pricing.
Studio is opening to research teams in waves and is free during beta. The seed therefore funds both product expansion and the transition from technical validation to a repeatable commercial workflow. If teams can order narrow datasets around observed failures and measure the improvement, Physicl can sell into an ongoing testing and retraining loop rather than compete only as a one-off data provider.
Founder and CEO Alex de Vigan previously built 3D infrastructure for retail. Physicl is applying that experience to a different customer problem: producing controlled, labelled worlds for systems that must act reliably outside a demo.
