Engineering researchers develop AI system allowing robots to reconstruct 3D scenes

Graduate students and faculty from the School of Engineering and Applied Science created a system that allows robots to reconstruct scenes outside of their direct line of sight using artificial intelligence.

The HoloRadar system uses AI to process radio signals and create three-dimensional images that help robots “see around corners.” The technology, developed by computer and information science assistant professor Mingmin Zhao and two graduate students, is unique in its availability to navigate dark spaces. 

“In some sense, the challenge is similar to walking into a room full of mirrors,” Zitong Lan, a third-year electrical and systems engineering doctoral student and co-author of the HoloRadar paper, wrote in an Engineering School release. “You see many copies of the same object reflected in different places, and the hard part is figuring out where things really are.”

The system uses radio signals which have longer wavelengths than visible light. While the length of these signals are “property traditionally seen as a disadvantage,” Zhao’s team recognized that the longer wavelengths were beneficial for scanning corners. HoloRadar is also able to navigate areas with minimal light which allows it to work in “the kinds of environments robots actually operate in,” according to Zhao. 

HoloRadar uses a two stage process to first, interpret a radio signal and extract the location’s apparent and hidden geometry, and second, use spatial reasoning to define a space and “reconstruct the actual 3D scene” in front of it.

“This capability is essential to help robots and autonomous vehicles make safer decisions in real time,” Zhao wrote.

The system’s lack of reliance on optimal lighting could lead to improved “safety and performance” of driverless cars and robots operating in cluttered indoor settings, such as warehouses and factories.

In addition to indoor exploration, Zhao’s team plans to explore outdoor environments in the future. These environments consist of locations “such as intersections and urban streets,” where longer distances “introduce additional challenges.”

The outdoor environments are necessary to provide robots with a “more complete understanding of their surroundings,” according to Zhao.

In December 2025, Zhao presented a paper about HoloRadar at the 39th annual conference on Neural Information Processing Systems.