Alex de Vigan is pivoting from the generative AI boom to a new venture that aims to enhance robotic capabilities. His startup, Physicl, has secured $10 million in seed funding from Inovia Capital to assist world model designers and robot engineers in overcoming training obstacles through the use of realistic digital simulations.
De Vigan spent a decade building nfinite, a platform that enabled brands to create 3D product models for e-commerce. While the company eventually shifted focus toward generative imagery due to transformer-based AI models, de Vigan recognized that the underlying technology for faithfully rendering 3D environments could be repurposed to address a critical gap in AI: the ability to understand, interact with, and manipulate physical space.
"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 stated.
Researchers creating physical AI models often struggle with the low returns of training robots on video and the high costs of data collection. Digital simulations are considered a promising alternative, yet ensuring these virtual worlds accurately mimic real-world physics has proven challenging.
de Vigan believes Physicl’s pipeline can solve this by transforming unstructured, incomplete real-world data into high-fidelity physical representations that include correct geometry, materials, and physics, thereby making physical AI models more effective.
Physicl enters a competitive landscape of startups providing data and tools for foundation model builders. The company plans to focus specifically on post-training models on difficult edge cases and has released a product called Studio, which includes 10,000 3D environments and 100,000 assets to help users generate synthetic data for specific robotic challenges, such as finding a cup in clutter or shadow.
SpAItial, a world model builder, is an early customer of Physicl and utilized the startup’s data to improve the fidelity of its latest product, Echo-3. SpAItial generates 3D simulations from 2D imagery and relies on Physicl’s high-quality synthetic data to do so effectively.
Physicl aims to be more than a simple data provider, positioning itself as a tool for ongoing synthetic data generation. The company envisions a workflow where massive amounts of unstructured data, such as trillions of images or LiDAR captures, are processed into high-quality 3D representations to train models on billions of permutations.
"The real challenge is, you have to curate the data by hand, and this is a very long cycle," said Matthias Niessner, CEO of SpAItial and a professor at the Technical University of Munich. He expressed excitement about Physicl’s potential to automate this process, allowing for on-the-fly data generation where systems like Claude can request specific training data as needed, a capability he views as highly innovative.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.