Safeworld launches lab to test robot safety in human simulations
The startup is building simulated human encounters to test robot control software, but its approach is still at an early stage.
Testing robots around simulated people
Safeworld emerged from stealth on October 5 to develop tests for robots that work around people, TechCrunch reported in an interview with its founders. The startup plans to place robot control software in simulated environments populated by human models and run many variations of an encounter. Co-founder Ding Zhao, who directs a Safe AI lab at Carnegie Mellon University, told TechCrunch that unpredictable behavior makes robot safety difficult to assess. The article describes examples such as a person appearing around a blind corner or falling near a machine.
A proposed method, not a certification
TechCrunch reported that the company raised more than $12 million in seed funding led by Shine Capital and a16z Speedrun. It also interviewed Gritt Robotics, which is working with Safeworld as the simulations are developed for robots used near solar-farm workers. The founders argue that robot makers may value outside testing, but the business is still deciding whether to offer a software platform or a service. No public results in the reviewed report establish that Safeworld’s simulations predict real-world safety or amount to a recognized certification standard. Its launch puts a concrete testing approach on the table; whether it reduces injuries will require evidence from deployments.