Carnegie Mellon Scholar Co-founds SafeWorld to Validate AI Robot Safety via Digital Twins

As generative AI is increasingly integrated into robotic control systems, the probabilistic nature of AI introduces new safety verification challenges. Ding Zhao, director of the Safe AI Lab at Carnegie Mellon University, co-founded SafeWorld alongside Kyle Wong and Simo Rachidi to develop safety simulation solutions for AI robotics. SafeWorld constructs digital twins of factory settings and visual blind spots in simulators such as Genesis and MuJoCo. They deploy software-driven robot models to test thousands of interaction scenarios with virtual human models, analyzing responses to events like stumbles and falls. Zhao noted that robots operate in unstructured environments with varying safety standards, making edge cases difficult to resolve. Vishal Dugar, CTO of partner firm Gritt Robotics, emphasized that safety guarantees require empirical verification that accounts for varied human postures and appearances. SafeWorld noted it is exploring platform and service-based business models and recently announced a $12 million seed funding round.

Key takeaways

• Carnegie Mellon scholar Ding Zhao co-founded SafeWorld to address the safety risks of generative AI-controlled robotics.
• SafeWorld uses Genesis and MuJoCo simulators to create digital twins, testing blind spots and human interaction edge cases like falls.
• Partner Gritt Robotics noted mathematical deduction alone cannot guarantee safety without empirical tests covering varied human postures.