A neural network designed to learn and clone car driving behavior in autonomous mode in Udacity Self Driving Car Simulator. The network is designed as a supervised regression problem between the car steering angles and road images from the car. The network is _ layers deep. The network predicts an angle for the car to manuever based on the input from camera in front of the car.
During training, images from three different camera angles i.e. from the center, the left and the right of the car are used.
src: The main project directory. The network and its training related functions are contained in model.py. The runtime inference and running a Flask app in the background are contained in drive.py.
data: You have to create this directory within this parent directory to store images for training from Udacity Self Driving Car Simulator.
model: Consists of a model that was training on the data that I collected and should drive the car around the simple and easy track without crashing. It also consists a visual representation of the network exported after training by model.py.
Python 3 and above, OpenCV 3 or above, NumPy, MoviePy, TensorFlow, Keras, Flask. There might be certain dependencies that may not be present in your environment. Hence, it is recommended that you setup a new virtual environment with help of requirements.txt
For training, the program assumes that training images and corresponding file with steering angles are present in "data"
python model.pyFor runtime inference, start drive.py first and then enter autonomous mode in the simulator.
python drive.py model.h5For a detailed explanation, please refer to my article on Medium.
Licensed under the MIT License.