Self-Driving Car Simulation

Abstract

Self-driving cars are one of the most favourable prospects of recent research in the field of Artificial Intelligence. The Futuristic view of this research anchors the large amounts of labelled and contextually rich data, which proliferate in driving. From complexity perception and control perspective, the technology to evidently solve driving can decisively be extended to other compelling tasks such as action recognition from videos and planning. An economically attractive approach for self-driving cars while still extending the AI frontier should be based on vision, which is equal to the main sensor used by a human driver. In this project, car driving behavior is cloned using neural network. The complication between the steering angles of the car and the images of the road in front of a car falls under supervised regression. Three different angles of the camera are considered (from the left, right and the center of the car) for capturing the images. The system is hinged on the proven NVIDIA model is a successful implementation in this problem field. The model is using convolutional layers for facilitating automated feature engineering involving image processing. Detecting useful road features with only the human steering angle as the training signal are the necessary processing steps for the internal representations which is automatically learned by the system.