Research on End to End Control Autonomous Driving System Based on Deep Learning

Abstract

Deep learning machine algorithm is successfully used in automatically driving of a toy
car. Firstly, the car was controlled by the system wirelessly to move along the
designated route, and the frame sample data and simultaneously records of the car
direction operation were collected using a camera to make the training data and label
separately. Secondly, the ResNet convolutional neural network was established by
using the frame sample data to predict the operation direction of the car. Thirdly, the
loss value calculated from the predicted action was compared with the value from
actual operation, and reduced by implementing the gradient descent algorithm. After a
series of experimental processes, the predicted value of the system can be equivalent to
the value obtained from actual driving action. The experimental results show that the
processing rate can effectively reach more than 30 fps with a high accuracy.