Physicl Raises $10M Seed Round to Fix Robot Training Simulations

Physicl has raised $10 million in seed funding to build more realistic digital simulations for training robots, according to TechCrunch. Inovia Capital led the round.

What problem is Physicl trying to solve?

Robot builders rely on simulated worlds to train physical AI models because gathering real-world data through remote operations is expensive, TechCrunch reports. But making those simulated worlds behave like the real one — matching true geometry, materials and physics — has been difficult. Physicl founder and CEO Alex de Vigan says that gap is the opening for his new company.

"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 told TechCrunch.

Where did this technology come from?

De Vigan spent the last decade running nfinite, a company that built 3D models of retail products for online shopping, according to TechCrunch. When generative AI models arrived, nfinite's core 3D-rendering niche shrank. De Vigan says the underlying pipeline — designed to turn "unstructured, incomplete, heterogeneous real-world information" into precise 3D representations — can now serve robotics instead of retail, per the TechCrunch report.

What is Physicl's new product?

The company is launching a tool called Studio. It ships with 10,000 pre-built 3D environments and 100,000 digital assets representing real-world objects, TechCrunch reports. Engineers can use Studio to assemble scenes and generate synthetic training data aimed at specific failure cases — for instance, teaching a robot to recognize a cup sitting in shadow on a cluttered counter, the outlet reports.

By the numbers

• $10 million — seed round led by Inovia Capital
• 10,000 — pre-built 3D environments in Studio
• 100,000 — digital assets available for scene-building

Who is already using it?

SpAItial, a world-model developer, worked with Physicl as an early customer to improve simulation fidelity, according to TechCrunch. SpAItial's own tool converts 2D imagery into 3D simulations; its latest release, Echo-3, came out last week, the outlet reports.

How crowded is this market?

Physicl is one of several startups selling data and tooling to companies building foundation models for robots and automation, TechCrunch reports. De Vigan acknowledged the competitive landscape directly. "The space is very young and very blurry sometimes," he told the outlet. Companies in the category are pursuing different approaches — collecting, labeling or generating training data in different formats — as customers search for the combination that makes robots reliable outside tightly controlled settings, according to TechCrunch.

Physicl's bet is narrower than some rivals: rather than building general-purpose world models, the company says it wants to focus on post-training existing models for hard edge cases, the kind of real-world clutter and lighting problems that simple simulations tend to miss.

AlterEgo — A keyboard for character voices. Five free persona rewrites a day.