Development of the End- to-End Learning based Autonomous Driving Framework and Experiments on a Full-Scale Autonomous Vehicle

In recent years, autonomous vehicles have been developed by various approaches for traffic safety and driver convenience. End-to-end learning-based autonomous driving has gained enormous attention in conjunction with deep learning technologies in which perception, planning, and control of the conventional autonomous driving algorithm are trained by a single deep neural network. In this paper, we present the end-to-end learning-based autonomous driving framework. The framework consisted of three parts: data acquisition in real-world and simulated environments, network design and optimization, and performance evaluation. Our framework was integrated on a full-scale autonomous vehicle platform and evaluated with various performance metrics.