Self-driving trucks, but make it video games
A Korean student team taught Euro Truck Simulator 2 to drive itself using OpenCV and TensorFlow—because real roads are expensive and virtual ones have reset buttons.

What it does ChosunTruck reads the game screen of Euro Truck Simulator 2, detects lane markings, and steers the truck to keep it centered. On Linux, an experimental TensorFlow-based mode can also spot other vehicles. It feeds keyboard and mouse inputs back into the game, so the truck physically drives itself while you watch.
The interesting bit The whole pipeline runs on screen capture and simulated input—no game modding, no memory hooks, just computer vision watching a window like a very focused human. The team picked ETS2 specifically because its roads look real enough to matter but forgiving enough to crash on.
Key highlights
- Lane detection works on Windows (Visual Studio build) and Linux (
make) - Optional vehicle detection on Linux via TensorFlow, enabled with
-Dflag - Hardcoded for 1024×768 windowed ETS2, 1920×1080 host display
- Requires manual window positioning on Linux—“right-down corner” is literal
- Open-source under GPLv3; actively seeking contributors
Caveats
- Windows setup is finicky: specific CUDA 7.5 install, read-only config hacks, exact sensitivity values in
controls.sii - Speed limited to 40–60 km/h on highways; not clear if it handles exits, traffic lights, or weather
- Future plans include switching from Tensorbox to YOLO2 and reading game memory directly—both still TODO
Verdict Grab it if you’re a student or hobbyist who wants to poke at autonomous driving without hardware budgets. Skip if you need something that works out of the box or handles more than gentle highway curves.