France’s Physicl Raises $10 Million for AI Training Data

PARIS -- Physicl, an artificial intelligence startup, has raised $10 million in seed funding to develop technology that generates synthetic training data for robots and other AI systems operating in the physical world.

Inovia Capital was named the lead investor during an announcement made Nov. 6, with the participation of U.S. Venture Partners. The money will go towards the growth of the company's team and improvement of its data creation platform.

Physicl has also unveiled a product called Studio, which allows scientists to create virtual 3D environments and generate data for particular weak spots in AI.

For instance, an AI that is taught to detect a cup on a transparent table would probably struggle doing so if the cup was partly covered or the light changed. Studio will help scientists create this situation virtually and create variations for better training.

Physicl claims that the platform gives access to 10,000 virtual 3D environments and more than 100,000 digital assets. The lighting, camera angles, and objects positioning can all be adjusted by researchers.

The company claims that the created dataset contains accurate information on both the objects and the environments, enabling comparisons between the models' predictions and known benchmark values.

 Physicl revealed partnerships with AI researchers such as Meta, World Labs, Epic Games, and SpAItial. The last company used Physicl data for training its models to construct a 3D representation based on pictures. It appears that using extra data reduced errors of the model's results by more than 50 percent, according to Physicl report.

 Nevertheless, this accomplishment applies to 3D reconstruction. Physicl did not share any similar achievements in the performance of real-world tasks by robots.