The big trend in robots is handing the keys over to a generative AI model, but that brings with it a problem: that architecture isn’t predictable the way traditional algorithms are. How can you be sure your brand new humanoid will be safe?
Dr. Ding Zhao, who directs the Safe AI lab at Carnegie Melon University, has been working on this problem for almost his entire career. Now, along with veteran start-up executive Kyle Wong and machine learning engineer Simo Rachidi, he’s founded a company, Safeworld, intended to solve it.
“The safety challenge that we’re talking about is a combination of, one, really advanced generative AI probabilistic evals — how do you underwrite the risk of a probabilistic system?” Zhao says. “The second part that’s really hard is the trust part, and you need both to deploy a robot.”
Safeworld is emerging from stealth today with a seed round of more than $12 million, led by Shine Capital and a16z Speedrun, with additional investment from Box Group, Carnegie Mellon University Endowment, Innovation Endeavors and SV Angel.
“The time to build an industry safety standard is now while robots are being designed and deployed,” a16z Speedrun partner Jonathan Lai told TechCurnch. “By the time you have robots in households colliding with kids and causing safety incidents, that’s way too late.”
Safeworld’s speciality is evaluating a robotic control system in simulations that are populated with realistic human models. It’s akin to the challenge faced by companies like Tesla or Wayve, who must ensure that their vehicles respond appropriately to a variety of surprising incidents they may encounter on the road. But that will be more difficult for robots, Zhao argues, because they work in unstructured environments, and because each facility they are in will have different safety standards.
“One of the most common areas is if there is a blind corner in this particular factory,” Wong said. “What is the speed or what is the stopping distance that you need to make sure that this robot will not collide with a particular human? If a human is carrying boxes, for example, will the robot detect the human or not?”
To answer that question, Safeworld will build a digital version of that corner in a model like Genesis or MuJoCo, insert a simulation of the robot it is evaluating, driven by its real software, and then run thousands of scenarios where human models encounter the robot. That’s harder than it seems, per Zhao, because people are unpredictable.
“Tripping and falling is also a good example of something that we do a lot of testing with the simulation,” Wong said. “Otherwise, you would have to go and trip and fall for the robot, which is like a hard thing to be doing all the time.”
There are definite similarities between the platform that Safeworld is building and the tools being used internally by robot builders. The founders, however, believe that beyond their specific expertise, robot-makers will want a third-party to validate their work, if only to share information about safety cases between competitors.
“A lot of people are underestimating one how hard some of these edge cases are going to be to solve,” Zhao said. “It is not the robot in the vacuum, in the demo, that we are worried about. It is the robot that is deployed at scale, with people who potentially never operated a robot before.”
Vishal Dugar, the CTO of Gritt Robotics, is developing the AI brain for robots that currently help workers install photovoltaic panels at industrial-scale solar farms, and aspire to take on more complex construction tasks. His company is partnering with Safeworld as they develop their safety simulations.
“The difficulty with most of our systems is it’s very hard to formally prove it by doing some math, writing some equations, and saying yeah, the system is verified to be safe,” Dugar says. “It necessarily has to be done empirically.”
His robots operate alongside human workers, and ensuring that its robotic arm doesn’t hit them is obviously top of mind. To verify that in practice will requires considering all kinds of potential scenarios.
“Humans have many kinds of appearances,” Dugar points out. “Their bodies can be in different configurations. They could be kneeling, standing. They could be tripping and falling potentially. They could be crouching. They could be running. You have to respond to all these behaviors that humans could potentially exhibit on these sites, along with the variety of variations in human appearance, you know, clothes, size, shape, height, skin color, everything else.”
It’s still early days for both Safeworld and generative AI in robotics, and the company is still figuring out the best model for its product—a platform for external users, or a services based approach?—but the team is confident they are taking on the right problem.
“We’ll probably be the first profitable company in this field,” Zaho says. “Because if anyone wants to deploy, they need to pay us to handle the situation.”
