Physicl raises $10 million and launches simulated robot training scenes

A cup hidden by clutter or shadows can expose a weakness in a robot's training. Physicl, a company creating simulated training data for robots and other AI systems that interpret the physical world, has announced $10 million in early funding and introduced Studio. The app lets researchers recreate situations their models struggle with, then vary lighting, camera angles, and object positions.

The company says this makes unusual failures easier to reproduce and test repeatedly. It reports that a customer more than halved errors when reconstructing real rooms after adding Physicl-generated data to its training material. The business is built around a specific need: once a team discovers where its AI fails, it needs suitable examples to teach it what went wrong.