New AI Predicts Whether a Driver Will Run a Yellow Light

Running a red light or slamming on the brakes at an intersection is one of the most common ways crashes happen. A team of researchers wanted to teach computers to see those moments coming. So they put instruments in real cars and recorded 449 trips through traffic lights, tracking the vehicle's position with GPS accurate to about a centimeter — roughly the width of your fingernail — while also monitoring the driver's heart rate and asking how comfortable each stop felt.

The most interesting finding is almost embarrassingly simple. The single strongest clue to whether someone stops or goes is how much braking the situation demands. If stopping would require only gentle pressure, most people stop. If it would require a hard, uncomfortable slam, many drivers gamble and drive through. From the data, the researchers identified five distinct comfort levels that match how humans actually brake.

Their AI works in two steps. First it predicts the decision: stop or go. Then it generates a realistic braking motion, with built-in physics rules so the car never brakes harder than a human body could tolerate. Tested against older methods, it predicted braking force within about 0.05 g and stopping position within roughly two feet. Given just one snapshot at the moment the light turns yellow, it can also forecast how comfortable or jarring the driver's eventual stop will feel.

The point isn't to nag drivers. It's to give self-driving cars and driver-assist systems a way to read human intentions early — the difference between a car that brakes after you do and one that brakes with you. The researchers published their data and code openly, which is good news for anyone building on it. The honest catch: 449 runs is a small sample, likely a handful of drivers in similar conditions, so the model reflects those habits rather than everyone's.