Steering Control for Autonomous Vehicles Using PID Control with Gradient Descent Tuning and Behavioral Cloning

In this paper we implement and evaluate two ways of controlling the steering angle of an autonomous vehicle, PID control with manual tuning followed by gradient descent algorithm tuning-which is able to enhance the performance through self-adjusting the controller parameters-and using supervised machine learning through the end-to-end deep learning for self-driving car which implement Convolutional Neural Network (CNN) to predict the steering angle for a given instance of a track. The verification testing went through two phases: software simulation using python for first run testing and C++ for simulation followed by track testing with a vehicle prototype. The proposed PID steering control system exhibits more stable steering commands-less oscillations-which makes it better than CNN Behavioral cloning control model. However, CNN Behavioral Cloning model may show better results after many several hours of training.