Startup Builds Virtual Humans to Safety-Test Robots

Startup Safeworld has raised more than $12 million in a seed round led by Shine Capital and a16z Speedrun. Carnegie Mellon University's endowment fund also joined the deal. The company builds virtual humans that stumble, fall, get up, and run inside a simulation. In these scenes, it tests how a robot reacts to an unpredictable person before that robot ever meets a real one. Safeworld was founded by Ding Zhao, who runs the Safe AI Lab at Carnegie Mellon.

With coding agents, you can check the output line by line and roll back a bad line in seconds. You cannot roll back a robot's mistake next to a real person the same way. Safeworld is bringing that same testing discipline to the robotics industry at large. It builds a simulation populated with people who look real, and it finds out how a robot handles a random, chaotic situation before that robot steps into the real world.