Fundamentals and Methods of Machine and Deep Learning

16Simulation of Self-Driving Cars Using Deep Learning

Rahul M. K.*, Praveen L. Uppunda, Vinayaka Raju S., Sumukh B. and C. Gururaj

Department of Telecommunication Engineering, B.M.S College of Engineering, Bengaluru, India

Abstract

Self-driving cars have been a popular area of research for a few decades now. However, the prevalent methods have proven inflexible and hard to scale in more complex environments. For example, in developing countries, the roads are more chaotic and unstructured as compared to developed countries. Hence, the rule-based self-driving methodologies currently being used in developed countries cannot be applied to the roads in developing countries. Therefore, in this paper, a methodology of implementing self-driving is discussed which we propose that it will be better suited to more unstructured and chaotic environments.

As discussed, there are many approaches to solving self-driving, each with its advantages and disadvantages. In this paper, the concept of end-to-end learning with behavioral cloning applied for self-driving is discussed. We have experimented with two different neural network models, Convolutional Neural Network (CNN) and Multilayer Perceptron (MLP). Also, we have two different pre-processing pipelines, that is, with and without lane feature extraction. In the final result, the goal is to find the optimal combination of these methodologies to get the highest accuracy. It was found that the end-to-end learning CNN model without lane feature ...