【作者】 张博；

【导师】 毕淑慧；

【作者基本信息】 济南大学 ， 控制工程（专业学位）， 2022， 硕士

【摘要】 为使移动机器人能准确、快速地到达指定的位置,其导航性能起着举足轻重的作用。地图构建精度是保证移动机器人导航精度的关键。在导航过程中,准确的点云地图能使移动机器人更加清晰地明确自己当前所在位置,以便合理规划后续路径。但目前常用的单一地图构建算法仍存在各种问题,因而精确的地图构建算法就成为导航研究的重点之一。首先,针对室内移动机器人的地图构建问题,本文通过了解各种常用的激光雷达和惯性测量单元,分析各种型号的激光雷达和惯性组件在技术参数及价格上的优势与不足,最终选定以RPLiDAR A1和ADI 9轴惯性组件结合的方式搭建组合地图构建实验平台,为验证后续提出的地图构建算法性能分析打下基础。其次,移动机器人在移动过程中容易出现轻微横滚以及俯仰姿态现象。通过研究发现,即便是微小的姿态偏移也会影响地图构建精度。针对该问题,本文通过对激光雷达的姿态进行模型构建,利用惯性组件获取移动机器人的姿态信息,通过提出基于Grahamscan算法的点云矫正方案实现对移动机器人的姿态进行矫正,以获取更加精确的环境点云信息。再次,针对因移动机器人位置信息不准确而影响点云拼合精度的问题,本文在已获得的环境点云信息基础上,通过引入路标,使移动机器人在构建地图环境点云信息时能够更加精确地获得当前位置信息。进一步,本文通过代入路标当前的精确位置并改进最近点迭代算法(Iterative Closest Point,ICP),提出改进ICP-路标组合地图拼接算法,完成点云信息图的拼合,使拼合效果更精确。然后,为进一步提升点云信息图的精确度,本文采用滤波算法对点云信息进行精加工。在地图构建应用中,发现扩展卡尔曼算法(Extended Kalman Filter,EKF)虽然收敛速度快,但其估计精度依赖于噪声特征统计的精度,若噪声的描述与实际环境的差距过大,EKF的估计性能会迅速下降甚至发散。此外,EKF对实际环境中的系统噪声Q和测量噪声R也无法做到精确描述。因此,为了优化采集的点云信息,本文研究自适应扩展卡尔曼算法(Adaptive Extended Kalman Filter,AEKF)、迭代扩展卡尔曼算法(Iterative Extended Kalman Filter,IEKF)以及有限长单位冲激响应滤波器(Finite Impulse Response,FIR)并提出改进AEKF滤波算法。然后将改进AEKF滤波算法与IEKF滤波算法相结合,提出改进自适应迭代扩展卡尔曼滤波算法(Adaptive Iterative Extended Kalman Filter,AIEKF)。实验结果表明,改进AIEKF滤波算法的性能要优于传统的AIEKF、AEKF、IEKF以及FIR等滤波算法。最后,本文通过采用遗传算法对滤波数据进行融合,提出基于遗传算法的FIR-改进AIEKF数据融合技术。通过遗传算法融合两种滤波算法的优势,得到最优的地图边界优化数据。实验结果表明,通过遗传算法融合后的地图边界精度优于单一滤波器的处理效果。总之,本文通过IMU进行移动机器人的姿态矫正,克服移动机器人姿态影响点云精度的问题。通过提出的改进ICP-路标组合地图拼接算法,解决移动机器人位置信息影响点云拼合精度的问题。通过提出的改进AIEKF滤波算法,改善单一EKF滤波算法过度依赖过程和测量噪声精确描述的缺点,相较于传统的EKF、AEKF、IEKF以及FIR滤波算法,改进AIEKF滤波算法有效提升点云信息图的精度。通过基于遗传算法融合FIR-改进AIEKF算法,将不同滤波算法的优点相结合,进一步提升点云信息图的精度。更多还原

【Abstract】 In order to make the mobile robot reach the designated position accurately and quickly,its navigation performance plays an important role.The accuracy of map construction is the key to ensure the navigation accuracy of mobile robots.In the process of navigation,an accurate point cloud map can make the mobile robot more clearly identify its current location,so as to reasonably plan the follow-up path.However,there are still various problems in the commonly used single map construction algorithm,so the accurate map construction algorithm has become one of the focuses of navigation research.First of all,for the map construction problem of indoor mobile robots,this thesis analyzes the technical aspects of various types of LiDAR and IMU by understanding various commonly used LiDAR and IMU due to the advantages and disadvantages of parameters and price,the combination of RPLiDAR A1 and ADI 9-axis IMU was finally selected to build an experimental platform for combined map construction,which lays the foundation for verifying the performance analysis of the map construction algorithm proposed later.Secondly,the mobile robot is prone to slight roll and pitch attitude phenomena during the movement process.Through research,it is found that even a small attitude offset will have a relatively large impact on the accuracy of map construction.Aiming at this problem,this thesis builds the LiDAR attitude model,uses the IMU to obtain the attitude information of the mobile robot,and proposes a point cloud correction scheme based on the Graham-scan algorithm to correct the attitude of the mobile robot to obtain a more accurate environment point cloud information.Thirdly,in view of the problem that the accuracy of point cloud stitching is affected by the inaccurate position information of the mobile robot,in this thesis,based on the obtained environmental point cloud information,by introducing road signs,the mobile robot can obtain the current position information more accurately when constructing the map environment point cloud information.Further,this thesis proposes an improved ICP-Roadmark combined map stitching algorithm by substituting the current precise position of the roadmark and improving the Iterative Closest Point(ICP)algorithm to complete the stitching of the point cloud information map and make the stitching effect more accurate.Then,in order to further improve the accuracy of the point cloud information map,this thesis uses a filtering algorithm to finish the point cloud information.In the application of map construction,it is found that the EKF has a fast convergence speed,but its estimation accuracy depends on the accuracy of noise feature statistics.If the gap between the description of the noise and the actual environment is too large,the estimated performance of the EKF rapidly degrades or even diverges.In addition,the EKF cannot accurately describe the system noise Q and measurement noise R in the actual environment.Therefore,in order to optimize the collected point cloud information,this thesis studies the AEKF,the IEKF,the FIR and proposed an improved AEKF algorithm.Then the improved AEKF filtering algorithm is combined with the IEKF filtering algorithm,and an improved AIEKF is proposed.The experimental results show that the performance of the improved AIEKF filtering algorithm is better than the traditional filtering algorithms such as AIEKF,AEKF,IEKF and FIR.Finally,this thesis proposes a FIR-improved AIEKF data fusion technology based on genetic algorithm by using genetic algorithm to fuse the filtered data.By combining the advantages of the two filtering algorithms through the genetic algorithm,the optimal map boundary optimization data is obtained.The experimental results show that the accuracy of the map boundary fused by the genetic algorithm is better than the processing effect of a single filter.In short,this thesis uses the IMU to correct the attitude of the mobile robot to overcome the problem that the attitude of the mobile robot affects the accuracy of the point cloud.Through the proposed improved ICP-Roadmark combined map stitching algorithm,the problem that the position information of the mobile robot affects the accuracy of point cloud stitching is solved.Through the proposed improved AIEKF filtering algorithm,the shortcomings of single EKF filtering algorithm over-reliance on process and accurate description of measurement noise are improved.Compared with the traditional EKF,AEKF,IEKF and FIR filtering algorithms,the improved AIEKF filtering algorithm effectively improves the accuracy of the point cloud information map.By combining the FIR-improved AIEKF algorithm based on the genetic algorithm,the advantages of different filtering algorithms are combined to further improve the accuracy of the point cloud information map.更多还原