Safeworld raises $12M to stress-test gen-AI robots in simulation

Safeworld, a Carnegie Mellon spinout that stress-tests AI-driven robots against simulated humans before they ever meet real ones, has raised over $12 million in seed funding, TechCrunch reports. The round was led by Shine Capital and a16z Speedrun, with Box Group, the Carnegie Mellon University Endowment, Innovation Endeavors, and SV Angel joining.

Co-founder Dr. Ding Zhao directs CMU's Safe AI Lab and started the company with Kyle Wong, a veteran startup executive, and Simo Rachidi, a machine learning engineer. Their argument is that formal mathematical verification is not available for generative-AI robot controllers moving among people, so testing has to be empirical. Safeworld runs customer robot software inside digital twins built on Genesis and MuJoCo, injecting human models that vary in size, shape, height, skin color, and behavior: kneeling, standing, tripping, crouching, running.

"Tripping and falling is something we do a lot of testing with in simulation, otherwise you'd have to go trip and fall for the robot," Wong told TechCrunch. Factory blind corners are another frequent scenario. The named early customer is Gritt Robotics, whose CTO Vishal Dugar is building AI for robots that install solar panels at industrial-scale farms.

Zhao's read on the market is that peers are too confident. "A lot of people are underestimating how hard some of these edge cases are going to be to solve," he said. He also expects the business to hold margin: "We'll probably be the first profitable company in this field." On who pays, his line is blunt: "If anyone wants to deploy, they need to pay us."

Safeworld has not yet decided whether to sell as a platform or as a service.