3D map construction of coal mine roadway mobile robot based on integrated factor graph optimization
-
摘要: 煤矿井下移动机器人作业精度严重依赖于同步定位与建图 (SLAM) 技术的准确性。井下长直巷道存在特征缺失、光照条件差等问题，导致激光里程计和视觉里程计易失效，因而限制了传统SLAM方法在煤矿巷道的有效应用，且目前SLAM方法的研究主要聚焦于多传感融合建图方法，较少关注激光 SLAM 方法建图精度的提升。针对上述问题，面向移动机器人在煤矿巷道的建图需求，提出了一种基于集成式因子图优化的煤矿巷道移动机器人三维地图构建方法，采用前端构建和后端优化的策略，设计了前端点云配准模块和基于滤波、图优化的后端构建方法，使建图结果更准确、适应性更强。针对煤矿长直巷道环境退化导致三维激光点云配准精度低的问题，融合迭代最近点 (ICP) 和正态分布变换 (NDT)算法，兼顾点云几何特征和概率分布特征，设计了集成式前端点云配准模块，实现了点云的精确配准。针对三维激光 SLAM 后端优化问题，研究了基于位姿图和因子图优化的后端构建方法，构建了集成 ICP和NDT 相对位姿因子的因子图优化模型，以准确估计移动机器人位姿。分别利用公开数据集 KITTI和模拟巷道点云数据集对三维地图构建方法在不同工况下的性能进行了实验验证。公开数据集 KITTI上的实验结果表明：在全局一致性上，该方法与传统基于特征点匹配的A−LOAM方法和基于平面分割及特征点提取的LeGO−LOAM方法具有相似的性能，在建图局部精度上优于其他2种方法。模拟巷道点云数据集上的实验结果表明：该方法具有显著优势，通过因子图优化，可得到一致性较高的三维地图，提升了煤矿巷道三维地图构建的精度及鲁棒性，解决了井下长直巷道特征点缺失、激光里程计失效的难题。Abstract: The working precision of mobile robots in coal mines seriously depends on the accuracy of simultaneous localization and mapping (SLAM) technology. There are some problems such as feature missing and poor lighting conditions in long and straight underground roadway. The problems lead to the failure of the laser odometer and visual odometer. The result limits the effective application of traditional SLAM method in coal mine roadway. At present, the research of the SLAM method mainly focuses on the multi-sensor fusion mapping method. There is a lack of research on the improvement of the mapping precision of the laser SLAM method. In order to solve the above problems, facing the mapping requirements of mobile robot in coal mine roadway, a 3D map construction method of coal mine roadway mobile robot based on integrated factor graph optimization is proposed. The method adopts the strategy of front-end construction and back-end optimization. The method designs a front-end point cloud registration module and a back-end construction method based on filtering and graph optimization. Therefore, the mapping result is more accurate and adaptable. The environmental degradation in coal mine long and straight roadway leads to the low registration precision of 3D laser point cloud. In order to solve the above problem, integrating iterative closest point (ICP) and normal-distributions transform (NDT) algorithms, taking into account the geometric characteristics and probability distribution characteristics of point clouds, an integrated front-end point cloud registration module is designed, which realizes the accurate registration of point clouds. Inview of the back-end optimization problem of 3D laser SLAM, the back-end construction method based on pose map and factor map optimization is studied. The factor map optimization model integrating ICP and NDT relative pose factors is constructed to accurately estimate the pose of the mobile robot. The performance of the proposed method of 3D map construction under different working conditions is verified by using the open dataset KITTI and the simulated roadway point cloud dataset. The experimental results on the open dataset KITTI show the following points. In terms of global consistency, this method has similar performance with the traditional A-LOAM method based on feature point matching and the LeGO-LOAM method based on plane segmentation and feature point extraction. It is superior to the other two methods in the local precision of mapping. The experimental results on the simulated roadway point cloud dataset show the following points. This method has significant advantages, through factor map optimization, a 3D map with high consistency can be obtained. The precision and robustness of 3D map construction of coal mine roadway are improved. The problems of the feature point missing and laser odometer failure in long straight underground roadway are solved.
-
图 7 INE−SLAM方法在模拟巷道的建图效果
Figure 7. Mapping effect of ICP and NDT ensemble SLAM in simulated roadway
图 8 LeGO−LOAM 方法在模拟巷道的建图结果
Figure 8. Mapping result of lightweight and ground-optimized LOAM in simulated roadway
图 10 巷道收窄部分INE−SLAM 方法的建图细节
Figure 10. Mapping details of ICP and NDT ensemble SLAM in narrowing roadway
表 1 APE对比
Table 1. Absolute pose error comparison
序列 方法 APE/m 最大值 平均值 中位数 最小值 均方根误差 和方差 标准偏差 A−LOAM 7.708074 2.014421 1.705119 0.291645 2.465952 27558.715930 1.422330 00 LeGO−LOAM 8.557403 4.429299 4.409769 0.351308 4.866022 107309.456000 2.014814 INE−SLAM 17.083902 7.400451 7.140376 0.808981 8.020236 291517.204500 3.091523 A−LOAM 35.109976 16.999973 18.074131 0.728562 18.893051 391928.228800 8.243076 01 LeGO−LOAM 290.476267 201.358024 224.237243 113.022321 210.274942 48548675.320000 60.584632 INE−SLAM 297.638420 203.560376 226.902531 112.478793 212.970114 49801183.740000 62.605453 A−LOAM 8.140921 1.971348 1.592937 0.228409 2.373178 15527.353030 1.321272 05 LeGO−LOAM 9.238267 2.139186 1.827800 0.482764 2.519107 17495.642090 1.330332 INE−SLAM 8.649879 2.256443 1.996284 0.472615 2.514627 17433.470720 1.109870 A−LOAM 3.285654 1.331722 1.012147 0.242015 1.545941 3795.214464 0.785143 09 LeGO−LOAM 6.063340 2.108868 1.971362 0.292446 2.343031 8717.794097 1.021015 INE−SLAM 7.315510 2.622855 2.616781 0.112678 2.840584 12813.436660 1.090663 
下载: 导出CSV
表 2 RPE对比
Table 2. Relative pose error comparison
序列 方法 RPE/m 最大值 平均值 中位数 最小值 均方根误差 和方差 标准偏差 A−LOAM 1.896651 1.154897 1.200041 0.003222 1.215805 6697.636647 0.379992 00 LeGO−LOAM 0.972212 0.061612 0.047549 0.003754 0.076631 26.607482 0.045567 INE−SLAM 0.514520 0.092929 0.078107 0.004084 0.109881 54.707031 0.058635 A−LOAM 3.885951 3.132486 3.553323 1.028025 3.218495 11363.507790 0.739085 01 LeGO−LOAM 2.699271 0.517959 0.169555 0.016541 0.831417 758.306416 0.650364 INE−SLAM 2.604537 0.517871 0.170409 0.020407 0.831250 758.001303 0.650220 A−LOAM 1.716019 1.123840 1.217291 0.000759 1.196405 3944.896806 0.410327 05 LeGO−LOAM 0.942990 0.053947 0.046318 0.002914 0.063128 10.983231 0.032786 INE−SLAM 0.883468 0.056378 0.047659 0.003852 0.065977 11.996595 0.034271 A−LOAM 2.197394 1.505677 1.491705 0.323694 1.550710 3816.262265 0.370999 09 LeGO−LOAM 0.334782 0.063748 0.057174 0.005735 0.070704 7.933544 0.030582 INE−SLAM 0.315042 0.064674 0.058037 0.008362 0.071649 8.146901 0.030836
下载: 导出CSV
-
[1] 杨林,马宏伟,王岩,等. 煤矿巡检机器人同步定位与 地图构建方法研究[J]. 工矿自动化,2019,45(9):18-24.YANG Lin,MA Hongwei,WANG Yan,et al. Research on method of simultaneous localization and mapping of coal mine inspection robot[J]. Industry and Mine Automation,2019,45(9):18-24. [2] 孙继平,江嬴. 矿井车辆无人驾驶关键技术研究[J]. 工矿自动化,2022,48(5):1-5,31. doi: 10.13272/j.issn.1671-251x.17947SUN Jiping,JIANG Ying. Research on key technologies of mine unmanned vehicle[J]. Journal of Mine Automation,2022,48(5):1-5,31. doi: 10.13272/j.issn.1671-251x.17947 [3] 潘祥生,陈晓晶. 矿用智能巡检机器人关键技术研究[J]. 工矿自动化,2020,46(10):43-48. doi: 10.13272/j.issn.1671-251x.2020080042PAN Xiangsheng,CHEN Xiaojing. Research on key technologies of mine-used intelligent inspection robot[J]. Industry and Mine Automation,2020,46(10):43-48. doi: 10.13272/j.issn.1671-251x.2020080042 [4] 周治国,曹江微,邸顺帆. 3D 激光雷达 SLAM 算法综述[J]. 仪器仪表学报,2021,42(9):13-27.ZHOU Zhiguo,CAO Jiangwei,DI Shunfan. Overview of 3D Lidar SLAM algorithms[J]. Chinese Journal of Scientific Instrument,2021,42(9):13-27. [5] ZHANG Ji, SINGH S. LOAM: lidar odometry and mapping in real-time[C]. Robotics: Science and Systems Conference, Berkeley, 2014. DOI: 10.15607/RSS. 2014.X.007. [6] HESS W, KOHLER D, RAPP H, et al. Real-time loop closure in 2D LIDAR SLAM[C]. IEEE International Conference on Robotics and Automation （ICRA） , Stockholm, 2016: 1271-1278. [7] SHAN Tixiao, ENGLOT B. LeGO-LOAM: lightweight and ground-optimized lidar odometry and mapping on variable terrain[C]. IEEE/RSJ International Conference on Intelligent Robots and Systems （IROS）, Madrid, 2018: 4758-4765. [8] SHAN Tixiao, ENGLOT B, MEYERS D, et al. LIO-SAM: tightly-coupled lidar inertial odometry via smoothing and mapping[C]. IEEE/RSJ International Conference on Intelligent Robots and Systems （IROS）, Madrid, 2020: 3106-3111. [9] BESL P J,MCKAY N D. A method for registration of 3D shapes[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1992,14(2):239-256. doi: 10.1109/34.121791 [10] LI Qingde ,GRIFFITHS J G. Iterative closest geometric objects registering[J]. Computers & Mathematics with Applications,2000,40(10/11):1171-1188. [11] CHEN Hui, BHANU B. Contour matching for 3D ear recognition[C]. 7th IEEE Workshop on Applications of Computer Vision, Breckenridge, 2005: 123-128. [12] HE Shijun, ZHAO Shiting, BAI Fan, et al. A method for spatial data registration based on PCA-ICP algorithm[C]. 3rd International Conference on Advanced Measurement and Test, Xiamen, 2013: 1063-1066. [13] BIBER P, STRASSER W. The normal distributions transform: a new approach to laser scan matching[C]. IEEE/RSJ International Conference on Intelligent Robots and Systems （IROS）, Las Vegas, 2003: 2743-2748. [14] KOIDE K,MIURA J,MENEGATTI E. A portable three-dimensional LIDAR-based system for long-term and wide-area people behavior measurement[J]. International Journal of Advanced Robotic Systems,2019,16(2):1729-8806. [15] LI Menggang,ZHU Hua,YOU Shaoze,et al. Efficient laser-based 3D SLAM for coal mine rescue robots[J]. IEEE Access,2019,7:14124-14138. doi: 10.1109/ACCESS.2018.2889304 [16] XU Jiachang,HUANG Yourui,ZHAO Ruijuan,et al. Probabilistic membrane computing-based SLAM for patrol UAVs in coal mines[J]. Journal of Sensors,2021:1-11. DOI: 10.1155/2021/7610126. [17] 王佳婧,王晓南,郑顺义,等. 三维点云初始配准方法的比较分析[J]. 测绘科学,2018,43(2):16-23. doi: 10.16251/j.cnki.1009-2307.2018.02.004WANG Jiajing,WANG Xiaonan,ZHENG Shunyi,et al. Comparison and analysis of initial registration methods of 3D point cloud[J]. Science of Surveying and Mapping,2018,43(2):16-23. doi: 10.16251/j.cnki.1009-2307.2018.02.004 [18] 李猛钢. 煤矿救援机器人导航系统研究[D]. 徐州: 中国矿业大学, 2017.LI Menggang. Research on navigation system of coal mine rescue robot [D]. Xuzhou: China University of Mining and Technology, 2017. [19] SCHULZ C, ZELL A. Real-time graph-based SLAM with occupancy normal distributions transforms [C]. IEEE International Conference on Robotics and Automation （ICRA）, Paris, 2020: 3105-3111. [20] PARK C, MOGHADAM P, KIM S, et al. Elastic LiDAR fusion: dense map-centric continuous-time SLAM[C]. IEEE International Conference on Robotics and Automation （ICRA）, Brisbane, 2018: 1-25. [21] QIN Chao, YE Haoyang, PRANATA C E, et al. LINS: a lidar-inertial state estimator for robust and efficient navigation[C]. IEEE International Conference on Robotics and Automation （ICRA）, Paris, 2020: 8899-8906. [22] KSCHISCHANG F,FREY B,LOELIGER H. Factor graphs and the sum-product algorithm[J]. IEEE Transactions on Information Theory,2001,47(2):498-498. doi: 10.1109/18.910572 [23] KUMMERLE R, GRISETTI G, STRASDAT H, et al. G2o: a general framework for graph optimization[C]. IEEE International Conference on Robotics and Automation, Shanghai, 2011: 3607-3613. [24] KAESS M,DELLAERT F. iSAM:incremental smoothing and mapping[J]. IEEE Transactions on Robotics,2008,24(6):1365-1378. [25] KAESS M, JOHANNSSON H, ROBERTS R, et al. iSAM2: incremental smoothing and mapping with fluid relinearization and incremental variable reordering[C]. IEEE International Conference on Robotics and Automation, Shanghai, 2011. DOI: 10.1109/ICRA.2011.5979641. [26] DELLAERT F. Factor graphs and GTSAM: a hands-on introduction [EB/OL]. （2016-01-13） [2022-09-16]. https://www.docin.com/p-1444792601.html. [27] GEIGER A,LENZ P,STILLER C,et al. Vision meets robotics:the KITTI dataset[J]. The International Journal of Robotics Research,2013,32(11):1231-1237. doi: 10.1177/0278364913491297 -
点击查看大图
图(10) / 表(2)
计量
- 文章访问数: 357
- HTML全文浏览量: 198
- PDF下载量: 45
- 被引次数: 0
