Jared Jones/Rice University
EyeDAR
The device is 3D printed and uses a lens that mimics the way the human eye focuses light, allowing self-driving cars to operate even when it is foggy or rainy.
After more than a decade of investments and billions of dollars invested in technology for autonomous vehicles, the bad weather continues to be an obstacle persistent. Heavy rain, fog and road spray continue to interfere with the sophisticated sensor systems used by self-driving cars, occasionally forcing vehicles to slow down, disable the system or even stop on the side of the road.
LiDAR cameras and laser systems generate high-resolution 3D maps, depend on light to function. Fog scatters light, rain distorts it and excessive glare can dazzle it, making driving difficult.
Now, engineers at Rice University and the University at Buffalo say they have developed a low-cost way to solve this blind spot. Its prototype, called EyeDARwas presented at HotMobile ’26 and focuses on a 3D printed component that costs around 6 euros to manufacture.
Unlike on-board sensors, EyeDAR is designed to be mounted on the side of the roadsimilar to a traffic sign. The device captures radar signals emitted by moving vehicles and determines the direction of origin of these signals. It then sends this directional data back, subtly modulating the vehicle’s own radar signal, eliminating the need for a separate transmitter, explains .
At the core of the device is a Luneburg lens, a spherical structure that bends incoming radar waves and focuses them onto a curved surface lined with antennas. The design mirrors the way the human eye focuses light in the retina. By physically focusing signals, EyeDAR avoids the high computational burden typical of conventional radar direction-finding systems and operates on just milliwatts of power.
In laboratory tests using a 24 GHz commercial automotive radar, the system demonstrated a high directional precision. A basic signal reading method had an average error of 0.68 degrees, while a refined weighted approach reduced the average error to almost zero within its effective angular resolution of 5.4 degrees. The device also maintained stable signal gain across the entire range of angles and transmitted data back to the radar with sufficient contrast for reliable detection.
The prototype does not yet measure distance, and a complete unit for use on public roads would require additional electronic components and a protective housing. Still, the researchers argue that a drastically lower price could make widespread infrastructure support economically viable.
