Safeworld raises $12m to safety-test robots controlled by generative AI

[Karen88]

Safeworld, a startup focused on the safety of robots controlled by generative AI, has emerged from stealth with 12 million dollars in seed funding, TechCrunch reports. It was founded by Carnegie Mellon's Dr Ding Zhao, executive Kyle Wong and engineer Simo Rachidi. Investors include Shine Capital, a16z Speedrun, Box Group, the Carnegie Mellon University endowment, Innovation Endeavors and SV Angel. Construction robotics company Gritt Robotics is a partner

The platform simulates how robots interact with digital models of people, so safety can be tested before real world deployment. It runs thousands of scenarios in physics engines such as Genesis and MuJoCo, including unpredictable human behaviour like tripping. The aim is to judge whether a robot can work safely alongside people before it is let loose in a warehouse, building site or home

Zhao described the safety challenge as a combination of advanced probabilistic testing of generative AI and what he called "the trust part", adding that "you need both to deploy a robot." He was also confident about the business. "We'll probably be the first profitable company in this field. Because if anyone wants to deploy, they need to pay us," he said. That is either refreshing honesty or impressive confidence

Jonathan Lai of a16z Speedrun argued that "the time to build an industry safety standard is now while robots are being designed and deployed." Gritt Robotics chief technology officer Vishal Dugar said this kind of validation "necessarily has to be done empirically." With humanoid robots improving fast, including Boston Dynamics' new Atlas hand that we discussed recently, testing how they behave around people is becoming urgent

Would you trust a robot in your home if it had passed simulated safety tests? And should there be a legal safety standard for AI-controlled robots before they go on sale?

Can Safeworld convince people that GenAI robots won't hurt them? | TechCrunchtechcrunch.com