PREDICT STEERING ANGLES IN SELF-DRIVING CARS USING INNOVATION CONVOLUTIONAL NEURAL NETWORK

Abstract

Nowadays, artificial intelligence and deep learning have emerged as evidence of the industrial revolution 4.0. Convolutional Neural Network (CNN) is one of the most popular Deep Learning network models, capable of recognizing and classifying images with high accuracy, even better than humans in many cases. This model has been applied to large image processing systems as Facebook, Google or Amazon... In this paper, we focus on studying some advanced CNN network models (VGG-16), based on VGG-16 architecture, we build new model, by increasing network depth, interleaved kernel 3x3, 1x1 increasing number of convolutional blocks, using Exponential Linear Unit (ELU) activation function after each convolution layer. Apply a new model to predict steering angles in autonomous driving based on image data obtained from Udacity self-driving car simulation software. Evaluation, experimentation, and research results show that the steering angle prediction in new model is really effective.