Physicl raises USD $10 million to train physical AI

Physicl has raised USD $10 million in seed funding and launched Studio. The round was led by Inovia Capital.

Studio is designed to help physical AI researchers generate training data for the situations in which their models fail. The Paris-based company is targeting a problem in robotics and related systems: developers often struggle to gather enough real-world data to cover rare or difficult scenarios before deployment.

According to Physicl, Studio provides access to 10,000 ground-truth 3D environments and more than 100,000 assets. Researchers can select an environment or describe a scene, then adjust factors such as lighting, camera viewpoint and object placement to recreate the conditions that caused a model to underperform.

The process is intended to help teams reproduce edge cases at scale, generate new data around those weaknesses and test whether model changes improve results. The platform supports a continuous loop of recreating failures, trying fixes and measuring progress.

Physicl says it already works with AI labs and universities including Meta, World Labs, Epic Games, SpAItial and Technical University of Munich. It also counts AWS and NVIDIA among the organisations it works with.

Founder and Chief Executive Officer Alex de Vigan said the business was built to address a gap between strong demonstrations and reliable real-world performance.

"An impressive demo doesn't tell you how a model will perform in all the situations it could encounter," said Alex de Vigan, Founder and Chief Executive Officer of Physicl.

"Models fail on edge cases. When researchers find these weaknesses, they need a way to generate data that specifically addresses them and test whether the model improves. We built Physicl to make that process easy and repeatable," he said.

Targeted training

Physicl traces its origins to de Vigan's work in 3D infrastructure, simulation and machine learning over six years. The new funding will be used to expand the team and continue developing the platform as robotics and world models move closer to wider use.

Investors are backing the company on the view that physical AI still faces practical obstacles outside controlled environments. These systems need to work consistently across a wide range of real-world settings, but collecting enough data for every possible condition remains difficult.

"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. That's what makes this team and its approach so compelling to us," said Steve Woods, Partner at Inovia Capital.

Early users

Physicl pointed to its work with SpAItial, a developer of world models that generate explorable 3D environments, as an example of how the system can be used. According to the company, SpAItial had been dealing with high error rates when reconstructing real rooms, and data generated from Physicl's library reduced that error rate by more than half.

De Vigan said the product is aimed at the routine but difficult failure cases physical AI teams face.

"A home robot might be able to recognise a cup on a clear counter, but miss it when clutter hides half of it or when the sun sets and shadows change in the kitchen," he said.

"Physical AI teams face challenges like this every day. Studio helps teams recreate those situations, generate variations with precise ground-truth, and train directly against the conditions where their model has failed in the past," he said.

For smaller research groups, the ability to create custom training data without collecting it physically could reduce the time and cost needed to improve models. Physicl says its system gives researchers access to a large set of virtual environments and scene variations that can be tailored to a model's weak points.

"Physicl gives any researcher access to virtually any physical training data they can imagine. That changes what a small team can build in physical AI, and we saw it in our results right away," said Matthias Niessner, Chief Executive Officer of SpAItial and Professor of Computer Science at Technical University of Munich.