A convolutional neural network (CNN) approach is used to implement a self-driving car by predicting the steering wheel angle from input images taken by three front cameras located in the center, left and right of the car. The architecture of the model used was developed by NVIDIA for its autonomous navigation system called DAVE-2. The CNN is tested on the Udacity simulator.
For more information about the project check the documentation.
The dataset is excluded from the repository due to its size. Data for this task can be gathered with the Udacity simulator itself. Indeed, when the simulator is set to training mode, the car is controlled by the human through the keyboard, and frames and steering directions are stored to disk.
The code is structured as follows:
data_preprocessing.py: contains all the necessary functions for data preprocessing.data_visualization.ipynb: visualizing data before and after data augmentation, and after image preprocessing.cnn_model.py: creates the CNN model.main.py: training and saving the model.autonomous_driving.py: driving the car in autonomous mode.model.h5: contain a trained CNN model.
The requirements for this project can be installed by running:
pip install -r requirements.txt
Note: It is recommended to use a virtualenv or conda for the installation of the requirements.
Using the Udacity simulator, the autonomous_driving.py file, and the already trained model.h5, the car can be driven autonomously around the tracks by executing:
python autonomous_driving.py
The model was tested on both tracks. On the first track, where the training data was collected, the model was able to run smoothly, however it is a bit unstable as it makes many small turns between left and right. On the other hand, in track 2 we can see that it has a better behavior and manages to run completely with a very good performance, so we can conclude that the model generalizes in a good way.
