Design and Validation of Reversing Assistant Based on Extreme Learning Machine

Abstract

As an important function of the advanced driver assistance system (ADAS), the reversing assistant (RA) achieves trajectory retracing by applying accurate position estimation and tracking control. To overcome the problem of the modeling complexity in dead reckoning for the reversing assistant function, the heading angular rate is compensated by using the extreme learning machine (ELM) to improve the positioning accuracy. In addition, considering the time delay of the steering system, a tracking controller with a feed-forward of the recorded steering angle and a self-tuning PID feedback controller is designed based on the preview-and-following scheme. Vehicle experiments under various reversing scenarios prove that the proposed positioning method and tracking control scheme are effective, the overall lateral error is less than 10 cm, and the heading angle error is less than 1°, which meets the requirements of performance indicators.