Physicl raises $10M seed led by Inovia to build synthetic 3D worlds for training robots

Physicl, a startup that produces synthetic 3D data for robots and other physical AI systems, announced a $10 million seed round on 6 November. Inovia Capital led the round, and The Next Web reports that U.S. Venture Partners also took part. Physicl describes itself as a US company with offices in New York, San Francisco and Paris and a team of about thirty people.

From product images to robot training

Founder and chief executive Alex de Vigan previously built Nfinite, a Paris company that makes 3D models and images of products for online retailers. TechCrunch reports that he saw the same rendering technology could serve robotics, where developers struggle to collect enough examples of unusual situations. Physicl came out of stealth at Nvidia's GTC conference in March and says most of its current staff worked with de Vigan on 3D production systems over the past six years.

The company's first application, Studio, launches with roughly 10,000 3D environments and more than 100,000 objects. Users pick a scene, vary one factor such as camera position, layout, lighting or clutter, and export images and video with ground-truth labels including depth, surface normals, segmentation and camera pose. The aim, de Vigan told TechCrunch, is to help teams post-train models on the hard edge cases where they fail, such as a cup in shadow on a crowded counter, rather than to sell one-off datasets.

Customers and claims

Physicl's website shows logos of organisations it says it works with, among them Meta, Nvidia, World Labs, Epic Games and the Technical University of Munich, without detailing each relationship. The clearest example is SpAItial, a Munich world-model startup run by TUM professor Matthias Niessner, who also advises Physicl. According to The Next Web, SpAItial says its errors when reconstructing real rooms dropped by more than half after it added Physicl's data. That figure comes from the companies and has not been independently tested.

A crowded field

Studio is still in beta, and Physicl has not published pricing. The company plans to use the money to grow its team and develop the platform. It enters a busy market: many startups now sell data collection, labelling or generation tools to robotics and foundation-model developers, and de Vigan himself described the space to TechCrunch as young and still blurry. Whether synthetic scenes close the gap with real-world data at scale remains the open question for the whole category.