University researchers are examining new safety certifications for autonomous systems, such as self-driving cars and delivery robots, a Feb. 17 news release announced.
Autonomous systems, like self-driving cars, currently follow “static” safety certifications that function like a final exam, wrote Ufuk Topcu, director of the Autonomous Systems Group, in an email response. He said these systems are tested against a “fixed checklist” of challenges and are approved if they pass. According to Topcu, autonomous robots are designed to operate in unpredictable environments, and their safety certifications should test versatility instead.
Cockrell School of Engineering researchers are examining these static safety systems to make them more adaptable.
“We want systems to be continuously evaluated as they evolve, as their environments change, and as our expectations of what ‘safe’ means change,” wrote Topcu.
Some are skeptical of the safety of these autonomous vehicles, but the problem of autonomous safety is “for all intents and purposes … solved,” said David Fridovich-Keil, an assistant professor in the department of aerospace engineering and engineering mechanics. In comparison to where autonomous vehicle technology was 10 years ago, their safety systems have come a very long way, he said.
Autonomous systems have become increasingly popular in recent years. Waymo cars, Uber’s take on self-driving cars, were introduced to Austin in March 2025. There have been 51 reported traffic incidents involving Waymos in Austin since June 2025, with six resulting in minor injuries, according to Robotaxi Tracker, which tracks autonomous vehicle traffic reports using data from the National Highway Traffic Safety Administration.
“Core safety challenges have remained consistent: uncertainty, adaptation, and interaction with complex environments,” Topcu wrote.
Fridovich-Keil says the team may run into some challenges due to the open-ended nature of the project. He says the ideal result of the study would be discovering the limits of dynamic safety systems, which adapt with time.
“Dynamic certification is hard,” Fridovich-Keil said. “There’s a reason it’s not solved yet. … I think even what is possible is still an open question.”
According to Topcu, an average day working on the project consists of a mix of testing mathematical theories, conducting experiments that reflect real-world conditions or meetings with a wide array of researchers.
“A significant part of the work involves discussions across fields: engineers, computer scientists, human factors experts, and researchers in learning systems all contribute different perspectives,” Topcu wrote.
This project relies on student contributions as well as professionals, Topcu wrote. Student participants engage with the research through mathematical modeling, simulation building or testing safety methods.
“The most important resource we have is our students,” Topcu wrote. “Watching students grow professionally, take ownership of complex problems, and contribute original ideas is one of the most rewarding parts of this work.”
