Physicl, a startup founded by Alex de Vigan, has raised $10 million in seed funding led by Inovia Capital to help robot builders and world model designers train their systems using more realistic simulations. The company is betting that technology de Vigan's previous team built to render products for e-commerce can be redirected toward making robots more capable.
De Vigan spent the past decade building nfinite, which let brands and retailers create 3D models of goods for display online. Transformer-based AI models pushed that company toward generative imagery, but de Vigan saw a new use for the rendering pipeline his team had assembled. He said the 3D work his company did uniquely was going to be replaced by AI, yet AI still lacked the ability to understand, interact with, and manipulate space and objects.
Physicl is targeting a known obstacle in physical AI research. Teams training robots to handle varied, small objects have seen limited returns from video-based training and face high costs when gathering data through remote operations. Simulated digital worlds are viewed as a promising alternative, but keeping those worlds faithful to real-world physics has been difficult. De Vigan argues his prior company's pipeline, which turns unstructured and incomplete real-world information into high-fidelity representations with correct geometry, materials, and physics, can make physical AI models more effective.
The startup is launching a product called Studio, offering 10,000 3D environments and 100,000 assets representing various objects. Users can build scenes and generate synthetic data to train robots on specific problems, such as a robot struggling to distinguish a cup on a cluttered counter-top or a cup sitting in shadow. Physicl aims to focus on post-training models for difficult edge cases rather than serving only as a data provider.
SpAItial, a world model builder, worked with Physicl as an early customer to improve the fidelity of its products, including its Echo-3 release. SpAItial generates 3D simulations from 2D imagery, and its CEO and founder Matthias Niessner said the company needed high-fidelity synthetic data of the kind Physicl offers to train effectively. Niessner, also a professor of robotics at the Technical University of Munich, said curating data by hand makes for a long cycle, and he described automated, on-the-fly data generation driven by training needs as attractive.
Physicl enters a crowded field of startups supplying data and tooling to foundation model builders and robot operators. De Vigan acknowledged the space is very young and sometimes blurry, with many companies collecting, labeling, or generating data in different modalities as customers search for what will make robots live up to their promise. He described what his team created a decade ago as, unknowingly, an agentic workflow for turning unstructured information into high-fidelity 3D, and he envisions customers feeding in trillions of images or millions of hours of lidar capture to produce 3D representations and generate permutations for training.
More company and startup news from TechManNews.
