Road Identification using Convolutional Neural Network on Autonomous Electric Vehicle

Abstract

Research in the field of autonomous electric vehicle has growth rapidly since they can overcome traffic accidents due to human error. Currently, the method used to identify the road for an autonomous electric vehicle is not in realtime. Thus, this study proposed a method for the autonomous electric vehicle to follow a predetermined route by identifying the road using the Convolutional Neural Network (CNN) as input of the steering control system. The optimal CNN model was obtained using an optimizer of Stochastic Gradient Descent with 150 epoch optimizer that was then used in simulation testing and real-time testing. In simulation testing, from 15 trials conducted, the percentage of success was 93.333%. The success rate to transmit the data from the system to the tool in a real-time manner is 100%. In real-time testing, the autonomous electric vehicle was successfully able to follow the predetermined route accurately. However, the autonomous electric vehicle has not succeeded in avoiding the object in front of it due to the lack of precise steering mechanics and the lack of variation in training data from various conditions that may be passed by the autonomous electric vehicle.