谢 辉,刘煜光,闫 龙
AuthorsHTML:谢 辉，刘煜光，闫 龙
AuthorsListE:Xie Hui,Liu Yuguang,Yan Long
AuthorsHTMLE:Xie Hui，Liu Yuguang，Yan Long
Unit:天津大学机械工程学院，天津 300072
Unit_EngLish:School of Mechanical Engineering，Tianjin University，Tianjin 300072，China
Abstract_Chinese:解决智能碾压机控制精度不达标的难题之一在于位姿信息的精确感知．为了改善智能碾压机在复杂环境下 的位姿感知效果，解决现有位姿感知算法存在精确性和普适性较差的问题，本文提出了用于智能碾压机位置测量和 航向测量的联合算法框架．首先，为了改善智能碾压机在复杂环境下的定位精确性，基于姿态传感器(AHRS)的加 速度计和陀螺仪测量值构建运动学预测方程，基于 GPS 的测量值和一组速度运动约束关系构建测量方程，在此基础 上设计基于 EKF 的松耦合算法对东向位置和北向位置进行精确输出．其次，为了解决碾压机铰接式结构带来的前后 车身航向角不一致的问题，通过分析 AHRS 输出磁方位角的原理以及输出数据构建航向角估计的经验方程式，并通 过反向传播(BP)神经网络模型进行变量值标定，对后车身航向角进行有效输出．最后，基于改装的智能碾压机平台 对位姿感知算法进行实验分析．实验结果表明：智能碾压机在路面波动下依靠 GPS 定位出现定位偏差时，该算法有 效地将偏差量进行补偿；在 GPS 短期失效 3 s 时，该算法输出坐标值与真实值的均方根误差为 10.9 cm；在低速状态 下，该算法输出航向角与 GPS 测量值的最大误差为 0.45°，有效解决了后车身航向角测量问题；说明位姿感知的联 合算法框架展现了较好的精确性和有效性．
Abstract_English:The accurate perception of pose information can help in solving the problem of the substandard control accuracy of self-driving rollers. This paper aims to improve the perception results of self-driving rollers in a complex environment and solve the problem of the poor accuracy and universality of the existing posture perception algorithms. To realize this，a joint algorithm framework for the position measurement and heading measurement of the self-driving roller is proposed. First，to improve the positioning accuracy of the self-driving roller in complex environments ， a kinematics prediction equation is constructed based on the attitude and heading reference system(AHRS)accelerometer data and gyroscope data，and GPS data and a set of constraint matrices of velocity are used to construct a measurement equation. Based on prediction equation and measurement equation，an extended Kalman filter-based loosely coupled algorithm is designed to calculate more accurate coordinates. Second，to solve the inconsistency of the front and rear bodies heading angle caused by the articulated structure of the roller，an empirical equation for estimating the heading angle is constructed by analyzing the principle of the AHRS output magnetic azimuth and the output data；then the variables in the equation are calibrated by the backpropagation(BP)neural network model. Finally，an experimental verification of the perception algorithm of the attitude and position is conducted on the modified intelligent roller platform. The experimental results show that when the self-driving roller relying only on GPS has measurement deviations due to road fluctuations，the algorithm effectively compensates for thedeviations. When the GPS failed for 3 s，the root-mean-square error between the output value of the system and the real value was 10.9 cm. At low speed，the maximum error between the output heading angle and the GPS measurement was 0.45°，which effectively solves the problem of measuring the heading angle of the rear body. The result shows that the perception system exhibits better accuracy and effectiveness.
Keyword_Chinese:智能碾压机；位姿感知算法；松耦合算法；航向角估计；BP 神经网络
Keywords_English:self-driving roller；perception algorithm of attitude and position；loosely coupled algorithm；estimation of heading angle；BP neural network

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6635