CNN based End to End Learning Steering Angle Prediction for Autonomous Electric Vehicle

Self driving vehicle is capable of sense the surrounding environment with the help of sensors and fulfill the transportation without any human control. It is a promising technology to reduce the accidents, increase the reliability and improve the transportation in our society. Steering angle prediction is an important feature for self driving vehicles. The normal approach for self driving vehicle is dividing into several steps such as road marking detection, vehicle detection, path planning and motor drive control. These human selected features are poor reliable to the high traffic environment and hence End to End learning method is implemented to overcome this drawback. This work explains the End to End learning method to predict the steering angle to keep up the self driving car in the lane by using the Convolutional Neural Network. The CNN architecture takes input images and predicts the steering angles consequently. This system is automatically learns inside representations of the imperative steps such as identifying beneficial road features with the driving steering angle along with road images as the training input. This work shows the ability of the convolutional neural networks to control the self driving vehicle by processing the input images. This work achieved a test loss of 0.0354 and R2 score of 0.819 using CNN architecture with different hyper parameters.