Driving Cars When Weather Gets Bad
Thu May 21 2026
The new system, called E2ETrADS, is built to help self‑driving cars keep going when the road gets slippery or dark.
Instead of relying on a handful of separate modules, it uses a transformer model that learns to drive by watching an expert.
The experts are drivers who use a smart planner and simple speed controls that change with the weather.
To teach the model, researchers ran many virtual drives in a simulator called CARLA.
They created scenes with clear skies as well as snow, rain and fog.
Every time the simulated car moved, a human‑like controller guided it, and the transformer learned from those actions.
When tested on new weather scenes, E2ETrADS did better than a previous benchmark called TransFuser.
It made fewer mistakes, stayed in its lane more often and changed speed smoothly even when the cameras were blurry.
The system also handled tricky turns that many other models missed, showing it can adapt to sudden sensor failures.
Beyond everyday traffic, the model survived rare but dangerous situations.
It showed that it could reason like a human driver in emergency scenarios, which is promising for future safety tests.
Overall, E2ETrADS demonstrates that a single deep‑learning network can learn to drive safely when the road is wet, snowy or dimly lit.
It keeps control stable and makes smarter decisions without needing separate weather‑specific modules.
