Alex de Vigan watched his previous company get swept away by the generative AI wave, but he's now wagering that the technology his team developed—a platform for crafting compelling product visuals for online retail—can be repurposed to make robots more capable. His new company, Physicl, has secured $10 million in seed funding from Inovia Capital to assist world model developers and robotics companies in tackling difficult training challenges through more lifelike simulations.

De Vigan, who serves as founder and CEO, spent the past ten years building nfinite, a company that simplified the process for brands and retailers to generate 3D models of their products and showcase them digitally. When transformer-based AI models emerged, that company shifted its focus toward generative imagery. However, de Vigan recognized a fresh application for his team and the technological foundation they had created for accurately rendering 3D environments.

"What we were super unique about doing in 3D, AI was going to replace, but something was missing in AI: its ability to understand, interact, and manipulate space and objects," de Vigan explains.

Researchers working on physical AI models—designed to help robots handle tasks like manipulating varied, small objects—have grown frustrated with the limited returns from training on video data and the high costs of collecting data through remote operations. Simulations in digital environments appear to be a promising answer, but making sure these virtual worlds accurately reflect real-world physics has proven challenging.

De Vigan is confident that the pipeline his former company built to "transform unstructured, incomplete, heterogeneous real-world information into a high-fidelity physical representation… with correct geometry, materials, and physics" can now be leveraged to enhance the effectiveness of physical AI models.

Physicl is stepping into a competitive landscape of startups all aiming to supply data and tools to foundation model developers and robot operators. "The space is very young and and and very blurry sometimes," de Vigan acknowledges. Numerous companies are gathering, annotating, or producing data across different formats as their clients search for the key ingredient that will help robots deliver on their potential.

Physicl specifically aims to concentrate on post-training models for challenging edge cases. The startup is launching a product called Studio, featuring 10,000 3D environments and 100,000 assets representing various objects. It enables users to construct scenes and generate synthetic data for training robots on particular problems. For instance, Studio could produce training data to assist a robot struggling to identify a cup on a messy countertop, or when the cup is in shadow.

EMBED_PLACEHOLDER_0

SpAItial, a world model developer, worked with Physicl as an early customer to enhance the accuracy of its products, the most recent being Echo-3, released last week. SpAItial's tool creates 3D simulations from 2D images, and to do so effectively, its CEO and founder Matthias Niessner said they required training on the type of high-fidelity synthetic data that Physicl provides.

Yet Physicl aspires to be more than just a data supplier; it positions itself as the toolkit for continuous synthetic data generation. "What we created 10 years ago, without knowing it, was basically an agentic workflow to transform unstructured information into high fidelity 3D," de Vigan said. He envisions customers bringing massive amounts of images or countless hours of LIDAR capture data into his platform, converting them into premium 3D representations that can generate billions of variations for training their models.

"To us what seems very attractive is you can just do this on the fly […] 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."