New Radar Trick Helps Self-Driving Cars See Through Rain and Fog

Most self-driving cars lean on cameras and lidar (laser sensors that map the world in 3D). Both struggle in rain, snow, fog and glare. Radar punches straight through that weather. A spinning radar sweeps a full circle around the car like a lighthouse beam, and newer versions can also record how fast each object is moving toward or away from it — the same Doppler effect that makes a passing ambulance siren change pitch.

The researchers asked a simple question: does that speed reading actually help the car recognise and follow vehicles? There was a catch. No public dataset had radar speed data alongside correctly labelled vehicles. So the team built one, using AI programs trained on lidar to automatically label 643 kilometres of real driving footage, then narrowed in on 250 km of busy, vehicle-heavy roads.

The results were encouraging. Feeding the speed data into the system cleaned up the radar picture, improving detection accuracy by up to 2.37 points. Even bigger: using measured speed to predict where a car will move next boosted tracking accuracy by 13.68 points compared with assuming every vehicle was standing still. Impressively, that came within 99.7% of the score you'd get from having perfect, hand-fed speed information.

So what? The takeaway is that radar's speed readings carry real, usable information — not just noise. A car that knows a truck is closing at 60 km/h can react earlier than one guessing from pictures alone. The catch: this is lab research, not a product. All the data came from one vehicle in one city, radar is still blurrier than lidar, and no carmaker has shipped it. But it points toward cheaper, weatherproof sensors doing more of the driving work — which could mean safer robotaxis sooner, and fewer sensor-heavy price tags.