As the robotics industry increasingly hands control over to generative SI models, a new company named Safeworld has emerged from stealth to address the resulting unpredictability. The startup aims to create safety evaluation tools for robotic systems that rely on probabilistic architectures rather than traditional deterministic algorithms.
What Happened
Safeworld was founded by Dr. Ding Zhao, director of the Safe SI lab at Carnegie Mellon University, alongside veteran startup executive Kyle Wong and machine learning engineer Simo Rachidi. The company announced 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 core challenge Safeworld addresses is the lack of predictability in generative SI-driven robots. "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 said. "The second part that’s really hard is the trust part, and you need both to deploy a robot."
Safeworld’s approach involves evaluating robotic control systems in simulations populated with realistic human models. Using platforms like Genesis or MuJoCo, the company builds digital versions of specific environments, such as a factory with a blind corner, and inserts simulations of the robot driven by its real software. They then run thousands of scenarios where human models interact with the robot to test for collisions and detection failures.
Why It Matters
The shift toward generative SI in robotics introduces significant safety hurdles. Unlike vehicles controlled by companies like Tesla or Wayve, which operate in structured road environments, robots often work in unstructured spaces where safety standards vary by facility. "The time to build an industry safety standard is now while robots are being designed and deployed," said Jonathan Lai, a partner at a16z Speedrun. "By the time you have robots in households colliding with kids and causing safety incidents, that’s way too late."
Vishal Dugar, CTO of Gritt Robotics, which uses SI brains for robots installing photovoltaic panels, partnered with Safeworld to develop safety simulations. Dugar noted that formal mathematical proof of safety is difficult for these systems, requiring empirical verification instead. "Humans have many kinds of appearances," Dugar said. "Their bodies can be in different configurations... 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."
The Bottom Line
Safeworld is still determining its final business model, whether a platform for external users or a services-based approach, but the founders believe the demand for third-party safety validation is critical for large-scale deployment. "We’ll probably be the first profitable company in this field," Zhao said. "Because if anyone wants to deploy, they need to pay us to handle the situation."