基于DBN的卡尔曼滤波算法研究及在AGV定位中的应用

发布时间：2018-10-19 15:31

【摘要】：AGV(Automated Guided Vehicles,自动导引车)是一种配备了自动导引装置,在没有人工指引或是驾驶的前提下,可以按照指定的路径行进,能够进行货物的装卸及搬运等一系列功能的自动化设备。AGV作为一种智能机器人,因其在货物搬运、目标牵引及生产线装配等领域的特殊用途,越发成为了各种工业生产中必不可少的重要工具。AGV以轮式移动作为主要的移动方式,它具有结构简单、快速高效、安全性及可控性良好等优点。总的来说,AGV系统需要三项重要技术:定位、路径规划和导引方式。其中定位技术作为AGV导航最重要的组成部分,是进行后续操作的基础。只有在准确地确定了当前的位置与姿态,才有可能根据生产过程中的实际需求,规划处一条有效行驶路径,为AGV完成后续任务做出详细的指引和保障。本文针对AGV中的定位技术展开相关研究。在常用的AGV设备中,经常使用卡尔曼滤波(Kalman filtering,KF)算法将传感器上获取到的信息进行融合,对AGV的位置和姿态做出估计。作为一种最优估计算法,KF算法可以对目标信息做出较为精确的估计。然而,使用KF算法的时候,存在一定的限制。如果不能预先确定噪声的统计特性,那么在AGV的运动过程中,KF算法难以对目标维持精确的估计。但是当外部环境复杂多变时,噪声的统计特性极易随外部的影响而发生改变,仅仅依靠KF算法是难以维持对位置和姿态的精确估计的。为了解决上述问题,本文将深度信念网络(Deep Belief Networks,DBN)和KF算法结合到一起。通过使用训练好的DBN模型产生调节因子,对噪声的协方差矩阵做出调整,从而维持KF算法对AGV的精确估计。最后,针对实际AGV的结构模型,建立AGV的运动学模型。使用KF算法对AGV的定位问题做出仿真,然后引入DBN对AGV的定位问题做出相应的仿真,并将两次的仿真结果作出分析。从分析结果可以看出,当噪声统计特性发生改变时,使用基于DBN的KF算法较之KF算法具有更好的鲁棒性,能够维持对AGV状态的精确估计。
[Abstract]:The AGV (Automated Guided Vehicles, (Automated guidance vehicle) is an automated guidance device that allows you to travel along a given path without manual guidance or driving. AGV, as an intelligent robot, has special applications in the fields of cargo handling, target traction and assembly of production lines. AGV takes wheeled movement as the main moving mode, which has the advantages of simple structure, fast and high efficiency, good safety and controllability and so on. In general, AGV systems require three important technologies: positioning, path planning, and guidance. As the most important part of AGV navigation, positioning technology is the basis of subsequent operation. Only when the current position and attitude are accurately determined can it be possible to plan an effective driving path according to the actual demand in the production process to provide detailed guidance and guarantee for AGV to complete its follow-up tasks. In this paper, the positioning technology in AGV is studied. In common AGV devices, Kalman filter (Kalman filtering,KF) algorithm is often used to fuse the information obtained from the sensor to estimate the position and attitude of AGV. As an optimal estimation algorithm, KF can estimate the target information accurately. However, there are some limitations when using KF algorithm. If the statistical characteristics of the noise can not be determined in advance, it is difficult for the KF algorithm to maintain an accurate estimation of the target in the process of AGV motion. However, when the external environment is complex and changeable, the statistical characteristics of noise can easily change with the external influence. It is difficult to maintain accurate estimation of position and attitude by using KF algorithm alone. In order to solve the above problems, this paper combines the deep belief network (Deep Belief Networks,DBN) and the KF algorithm. The covariance matrix of the noise is adjusted by using the trained DBN model to generate the adjustment factor, so as to maintain the accurate estimation of the AGV by the KF algorithm. Finally, the kinematics model of AGV is established according to the structure model of actual AGV. The localization problem of AGV is simulated by using KF algorithm, and then the corresponding simulation of AGV location problem is made by introducing DBN, and the results of twice simulation are analyzed. It can be seen from the analysis results that the KF algorithm based on DBN is more robust than the KF algorithm when the statistical characteristics of the noise change, and it can maintain the accurate estimation of the AGV state.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP23;TN713

【相似文献】