基于动态步长梯度下降的EKF-SLAM改进算法  

Improved EKF-SLAM Algorithm Based on Dynamic Step-size Gradient Descent

作　　者：索会恒 魏博思 饶睿 胡强 钟璟 杨腾胜 吴剑[1] SUO Huiheng;WEI Bosi;RAO Rui;HU Qiang;ZHONG Jing;YANG Tengsheng;WU Jian(School of Information Engineering,Nanchang Hangkong University,Nanchang 330063,China)

机构地区：[1]南昌航空大学信息工程学院,南昌330063

出　　处：《常熟理工学院学报》2024年第2期63-70,共8页Journal of Changshu Institute of Technology

基　　金：南昌航空大学研究生创新专项“ROS环境下的移动机器人实时定位导航系统设计”(YC2021-S679);南昌航空大学研究生创新专项“基于多智能体的多机器人协同任务规划技术研究”(YC2021-037)。

摘　　要：现代重型货物运输所处环境日益复杂,要求AGV-SLAM算法具备更高的鲁棒性和准确性以实现高精确装卸作业.因此,本文提出基于动态步长梯度下降的改进EKF-SLAM算法.首先对AGV行进过程建模分析,分别建立AGV的运动学模型和观测模型.考虑到EKF-SLAM算法在状态估计时可能引入的截断误差问题,本文针对AGV自身位姿和环境特征位置预测估计精确度较低的问题,提出了一种改进算法.该算法引入梯度下降算法,并结合动态步长策略,步长的取值与AGV系统的前进速度、采样时间正相关.仿真结果表明,该改进算法相较于标准EKF-SLAM算法,在停车场数据集下能够快速生成更优的估计路标和路径,从而在一定程度上提高了传统算法的鲁棒性和准确性,具有行业应用的参考价值.The increasingly complex environment for modern heavy-duty goods transportation requires AGV-SLAM algorithms to have higher robustness and accuracy in order to achieve precise loading and unloading operations.Therefore,this paper proposes an improved EKF-SLAM algorithm based on dynamic step-size gradient descent.Firstly,the AGV motion process is modeled and analyzed,and the kinematic and observation models of the AGV are established.Considering the truncation error problem that may be introduced by the EKF-SLAM algorithm during state estimation,this paper proposes an improved algorithm to address the issue of low accuracy in the prediction estimation of AGV's own pose and environmental feature positions.The improved algorithm introduces gradient descent algorithm and dynamic step-size strategy,where the step-size is positively correlated with the AGV's forward speed and sampling time.Simulation results show that compared with the standard EKF-SLAM algorithm,the improved algorithm can quickly generate better-estimated landmarks and paths in the parking lot data set,thus improving the robustness and accuracy of traditional algorithms to a certain extent,and carries reference value for applications in the industry.

关 键 词：同时定位与地图创建 动态步长 扩展卡尔曼滤波 停车场数据集 AGV 

分 类 号：P237[天文地球—摄影测量与遥感] TP2[天文地球—测绘科学与技术]