Self-Driving RC Car

This project is a part of my bachelor's thesis, where we tackled the complex problem of building autonomous vehicles by creating a small-scale solution. Utilizing Convolutional Neural Networks (CNNs), we predicted the car's directional movement based on real-time track images. We captured these images using a Raspberry Pi and a Pi camera module and communicated with a local server through TCP/IP socket programming.

Here's the setup of the RC car:

Track set up

Dataset

MAC0460 Self driving dataset available on Kaggle is used for training the classifier.

Model Training

The training involves multiple steps:

1. time_stamps_match: To synchronize Image_Stamps and Direction_Stamps, generating Image_Command.csv.
2. training_file_gen.ipynb: To create Training_Data and Training_Labels from Image_Command.csv.

Models

• ANN: neural_network_simulation from scratch.
• CNN: RC_CNN.py (Training accuracy up to 97.64%)

For manual Data labeling

• Driving with Arduino and Keyboard: rc_car_controller
• Driving with Raspberry Pi and Keyboard: rpi-car-controller

For Testing/Driving the car using Model Predictions

1. rc_rpi_controller: Arduino-based socket connection for car commands.
2. drive.py: RPI program to actuate car based on model's predictions.

For in-depth information and model training/testing results, please refer to our project report.