A pattern as ordinary as stripes on a fence could have serious implications for the safety of autonomous vehicles, drones and robots. New research from UF Herbert Wertheim College of Engineering researchers has uncovered a fundamental weakness in the way these systems perceive depth, showing that repeated visual patterns can cause an obstacle to appear significantly closer or farther away than it really is.
The research, led by Sara Rampazzi, assistant professor in the department of computer and information science and engineering, goes beyond identifying the vulnerability. Her team tested the problem across multiple sensors, algorithms and AI models and developed defenses aimed at addressing the underlying cause. The findings highlight the importance of building autonomous systems that can reliably interpret the physical world, even when unexpected visual conditions interfere with their perception.