Self-Driving Cars: Exploring the Potential of Using Convolutional Neural Network to Overcome Road Variation
Abstract/Artist's Statement
The use of self-driving cars can benefit the society in many ways, such as reducing traffic accidents and enabling disabled people to travel independently. The potential of reducing traffic accidents can be considered most important, since in 2017, mistakes made by human drivers were the cause of over 90% of the traffic accidents, leading to 40,100 people’s deaths in the United States. If human drivers were replaced by autonomous systems, the number of traffic accidents would decrease. Although the concept of self-driving car was raised since at least the 1920s, a commonly accepted development of self-driving car has not yet appeared. A significant challenge is the creation of a system that can accurately detect the environment around itself and then form the right driving command. Recent progress in deep learning suggested that convolutional neural networks are a form of machine learning that can be trained to extract features and use those features to control a car. This project focuses on extending the network model in the paper published by NVIDA in 2016 [1]. The aim of the project is to evaluate how well a convolutional neural network could perform on a simple, simulated roadway with road varying and missing road edges.