End-to-End Deep Learning for Autonomous Longitudinal and Lateral Control based on Vehicle Dynamics

Abstract

An end to end method predicting decisions by using deep learning method to mimic driving behaviors from observed images information is one of the famous methods for developing an autonomous self-driving car. In this paper, we investigate the end to end method based on the deep convolution neural network by considering the vehicle dynamic to mimic decisions of human drivers such as steering angle, acceleration, and deceleration. The effect due to the vehicle dynamics of host car by ignoring previous states is investigated through the comparison of predicted accurate and variation by collecting real data in a simulation study.