Lane following using behavioural cloning

Abstract

With the rise in the research relating to Artificial Intelligence along with the growing concern of everyday road accidents due to human error, the research pertaining to Autonomous Vehicles has been soaring to new highs. However, this technology in its current form has serious limitations such as restricted use during adverse conditions (such as snow), inability to identify manual traffic instructions, abnormal traffic behaviours etc. This is one of the reasons that even the vehicles with most autonomous features, exhibit only a Level 2 or Level 3 of driving automation. Hence, in order to reach further levels of automation, it may be useful to create a symbiotic technology between autonomous vehicles and traffic control models. This thesis work will work as an elementary stepping stone to create such a symbiosis by identifying a Lane Following Model using Convolutional Neural Networks. Specifically, a Behavioural Cloning Model along with a Road Classification Model is developed in order to mimic human driving characteristics which ideally works independent of lane markings and to regulate this driving characteristics by reading road signs with satisfactory levels of accuracy.