LSTM 딥 러닝 기반의 비 모델 LKAS 조향 각 생성

Abstract

In this paper, we propose a lateral upper controller that predicts vehicle motion and generates a model-free LKAS steering angle by applying long-term memory (LSTM), one of the deep learning techniques. The apparent and distinct advantage of this LSTM model is that the relationship of nonlinear sensor data can be grasped and learned devoid of considering a mathematical model with the aim of time information. In this sense, there has an important fact from related to this. The LKAS steering angle can be generated by predicting the movement of the vehicle in consideration of the vehicle movement from past to present based on the sensor data of the vehicle, indicating the flow of it. In addition, In relation to time delay problem due to the difference of sampling time on each sensor, it can be simply solved by learning based on a time table which has a synchronized sampling time. The input values of the upper controller were selected as the coefficient values of the road model and the vehicle dynamic characteristics data came from the image data output through the camera sensor obtained from image processing, and therefore, the target values were selected as the steering angles in the next state of the current state. The learning model was developed by learning the many-to-one LSTM prediction model with serial connection of LSTM and Fully-Connected (FC) Multilayer Perceptron (MLP). After following this, test data were tested and compared with a model - based multi rate Kalman Filter (MKF) LKAS lateral controller. In setting a design, learning and testing, it was used Tensorflow, and the study was conducted with the data obtained by driving the road test course of the road traffic safety corporation.