This Robot Can See Around Corners

At a high level, autonomous vehicles all use sensors to detect what is in front of them. Then they process this data to determine what objects are present before an onboard algorithm uses that information to determine what action to take next. The latest self-driving systems that use some variation of this approach are quite reliable. But how much safer would they be if they could see beyond what is directly in front of them? What if they could see around corners?

That is what a team of engineers at Penn State University set out to determine. They built a system, called HoloRadar, that uses radio signals and AI to reconstruct three-dimensional scenes outside their direct line of sight, such as pedestrians around the corner. Unlike existing approaches, this new technique does not require visible light, so it works reliably even at night.

The system relies on a counterintuitive property of radio waves. Because radio wavelengths are much larger than the microscopic roughness of common surfaces like drywall and concrete, walls behave almost like mirrors for the signals. Instead of scattering randomly as light does, radio pulses bounce predictably off floors, ceilings, and corners, carrying information about hidden areas back to the robot’s sensor. In effect, the environment itself becomes a network of invisible mirrors.

Capturing those reflections is only the first step. A single pulse can ricochet multiple times before returning, creating a tangled mixture of echoes that traditional radar processing struggles to interpret. To solve this, the researchers developed a two-stage AI system. The first stage enhances the low-resolution radar readings and identifies multiple “returns,” estimating how far each reflection traveled. The second stage uses a physics-guided model to trace those paths backward, undoing the mirror effect and reconstructing where objects actually are in the real world.

The researchers describe the challenge as similar to standing in a room full of mirrors and trying to figure out which reflections correspond to the real object. By explicitly modeling how radio waves interact with surfaces, the software can separate genuine positions from mirrored illusions and produce a full 3D map — not just detecting a hidden person, but reconstructing hallways, walls, and obstacles around a corner.

In tests on a mobile robot navigating real building corridors and corners, the system successfully reconstructed both visible and hidden spaces, including human subjects outside the robot’s line of sight. Researchers now plan to extend the work outdoors, where longer distances and constantly moving objects present additional challenges.

If successful, the technology could transform how machines perceive the world — giving robots and vehicles a more complete awareness of their surroundings, and perhaps one day allowing them to anticipate danger before it ever comes into view.