Self-Driving RC Car with DonkeyCar and a Neural Network
Drive laps to collect data, train an AI model on your computer and let the car drive itself.
DonkeyCar is an open-source platform for small self-driving cars. You drive the track a few dozen laps, the car records camera images with your steering, and a neural network learns to copy you.
Run autonomous tests in a closed area with no people, pets or traffic.
Video tutorial
Video by ebswift, embedded from YouTube. The written steps below are our own and match the parts list.
Mission steps
Build the hardware
Use the smart RC car wiring (PCA9685 or a direct PWM setup) and mount a wide camera high and centered, angled toward the track.
Install DonkeyCar
Follow the official DonkeyCar docs for Raspberry Pi: create a Python environment, install the donkeycar package and run donkey createcar to make your car folder.
Calibrate
Use donkey calibrate to find steering and throttle limits, then save them in myconfig.py.
Collect training data
Mark a track with tape, start manage.py drive, steer with the web or game controller and record 10–20 clean laps.
Train and drive
Copy the data to a laptop with a GPU (or a cloud notebook), run donkey train, copy the model back and start the car in autopilot mode.
Questions
How much data do I need?
Roughly 5,000–20,000 images of smooth, consistent driving. Bad laps teach bad habits — delete them.