자율주행을 위한 공간적 경향성이 고려된 End-to-End Neural Network 설계

Abstract

Recently, autonomous driving and advanced driver assistance system (ADAS) have been actively researched in the automotive field. It is very important to consider the improvement of perception ability about a forward driving scene. However, the sensor"s capability maintains high quality when the driving condition is only ideal. There are so many road conditions and an environment in real driving situations. In this paper, we propose a steering wheel angle prediction model using a deep convolutional end-to-end neural network for various driving environments and road shapes. The image data for training and validation is UDACITY Challenge dataset and the experiment was performed through computational simulation.