Physicl, a Paris-based start-up that creates training data for AI systems that work in the physical world, has raised $10M in seed funding. Inovia Capital led the round, announced on Friday, with U.S. Venture Partners also taking part.

The company focuses on what researchers call edge cases. These are the unusual situations, such as odd lighting or an unexpected object in the way, where AI models for robots, self-driving cars, and 3D vision often make mistakes.

Its first product, Studio, lets researchers rebuild those situations in 3D. They pick an environment or describe a scene where their model struggles, then change the lighting, the camera angle, and where objects sit.

Studio then produces accurate training data aimed at that specific weakness. As a result, researchers can recreate a failure, test a fix, and measure whether their model has improved.

The platform includes 10,000 3D environments and more than 100,000 objects, which can be combined into billions of different scenes, according to the company.

“Models fail on edge cases. When researchers find these weaknesses, they need a way to generate data that specifically addresses them,” said Alex de Vigan, Physicl’s founder and chief executive.

De Vigan previously founded Nfinite, a Paris company that makes 3D images of products for online retailers and raised $100m in 2022. He has spent six years working on 3D, simulation, and machine learning.

Physicl already works with AI labs and universities, including Meta, World Labs, Epic Games, Amazon Web Services, Nvidia, and the Technical University of Munich.

One customer is SpAItial, a Munich start-up that builds 3D AI models. After adding Physicl’s data to its training set, SpAItial cut the error rate of its room-building model by more than half, the company said.

Physicl will use the capital to grow its team and develop the platform further. It joins a wave of European start-ups working on physical AI, the technology that lets machines understand and act in the real world.