Improving Car‑Following Models on Icy Roads with AI

The new study tackles how cars behave when roads are slick and visibility is low. It looks at five popular driving models, each with its own set of adjustable numbers that dictate how a vehicle follows another. The researchers first list the main variables that matter in winter: how much grip the road offers and how far a driver can see. Next, they point out that old calibration tools, like simple genetic searches, do not adapt well to these harsh conditions. To solve this, they mix two modern techniques. One part uses an Informer encoder that reads patterns in the weather data and turns them into useful signals. The other part uses a physics‑informed neural network that changes the model’s numbers on the fly, respecting real traffic laws.When tested with a well‑known highway dataset and real car runs in snow, the new method performs better. It improves the match between simulated and actual speeds by about twelve percent compared with the older method. Moreover, it works across all five models without needing a new tuning process each time. The research shows that combining data‑driven insights with physical rules can give traffic simulators a clearer picture of winter driving. This helps planners design safer roads and drivers better prepare for icy conditions.