New Radar Tech Lets Robots Drive Through Fog, Dust and Smoke

Most self-driving cars and delivery robots see the world with lidar — a spinning sensor that bounces laser light off nearby objects to build a 3D picture. It's precise, but it has a fatal weakness: fog, rain, dust and smoke scatter laser light, so the vehicle effectively goes blind in exactly the conditions where you'd most want a robot doing the driving. Radar, which uses radio waves instead, passes straight through those same particles.

The problem has been that radar pictures are blurrier and messier than laser pictures, so researchers historically threw away most of the data and used only a few strong points. This team did the opposite. Their system, called DRT&R (Direct Radar Teach & Repeat), keeps the entire radar scan and combines it with a local map built on the fly. The name describes the idea: the vehicle is "taught" a route once, then repeats it by matching what its radar currently sees against that map — no GPS required.

The results are striking. Across 344 km of on-road driving and 20 km of off-road terrain, the system usually pinpointed its location to within 4 cm — roughly the width of a golf ball. Even in sparse, featureless environments where there's almost nothing to navigate by, it stayed within 12 cm. And in the real test, a four-wheeled off-road robot called a Clearpath Warthog drove more than 10 km completely on its own, in real time, using the system to steer.

So what? Robotic delivery vans, mining trucks and farm equipment currently stop working when the weather turns bad. A navigation system that shrugs off dust and fog could keep them running through storms, in mines, and across wildfire smoke — and could one day make your own car's driver-assistance features more reliable in the rain.