Design of a Machine Learning-Based Self-driving Car

In this work, an algorithm of machine learning for self-driving car using udacity and unity self-driving car simulation software has been presented. Using the software, the car is driven on the simulated circuit having three cameras mounted on car hood which generate three images simultaneously and acceleration and de-acceleration of the car steering angle and brake. This method includes a behavior cloning approach and tries to replicate a behavior of human driver. For training the model, approximately, 18,000 training samples are required, and by using image augmentation technique, an increase in the data sample with few times is obtained, which leads to little robust simulated self-driving car.