基于全局路径优化的移动机器人3D激光融合定位与建图

Study on 3D laser information introspection of positioning and mapping for mobile robots based on global path optimization

摘要:

基于工厂车间无人化智能仓储的移动机器人自主定位和导航需求,设计了基于全局路径优化的3D激光融合同时定位与建图方法。前端里程计算法融合了激光雷达和IMU数据,采用ICP(迭代最近点)点云配准算法进行激光点云特征点匹配,通过初始定位流程及线段匹配技术,实现机器人在全局地图中的动态定位,包括惯性导航位姿推算,地图匹配的位姿计算和动态重定位,实现IMU与激光雷达的数据同步和机器人位姿的实时准确估计。同时定位与建图(SLAM)后端算法包括全局位姿优化算法,基于Scan Context(全局描述符)的回环检测,基于图优化的全局位姿优化等。通过考虑整个轨迹上的所有观测数据来优化机器人的位姿估计,基于Scan Context的回环检测通过比较不同时间点的Scan Context来确定机器人是否回到了之前访问过的位置。基于图优化的全局位姿优化通过构建位姿图,考虑里程计、回环检测和RTK数据作为约束,对全局位姿进行优化。最后通过无人叉车型移动机器人定位建图实验验证了所提出算法的定位精度和运行可靠性。

Abstract:

Based on the demand for autonomous positioning and navigation of mobile robots in unmanned intelligent warehousing of factory workshops, a 3D laser fusion simultaneous localization and mapping method based on global path optimization is designed. The front-end odometry calculation method fuses lidar and IMU data and employs the ICP (Iterative Closest Point) point cloud registration algorithm for laser point cloud feature point matching. The initial positioning process determines the position of the robot in the global map through line segment matching technology. Dynamic positioning includes inertial navigation pose estimation, map-matched pose calculation, and dynamic relocation, realizing data synchronization of IMU and lidar and real-time and accurate estimation of robot poses. The simultaneous localization and mapping (SLAM) back-end algorithm consists of a global pose optimization algorithm, loop closure detection based on Scan Context (global descriptor), and global pose optimization based on graph optimization. By considering all observation data on the entire trajectory, the pose estimation of the robot is optimized. Loop closure detection based on Scan Context determines whether the robot has returned to a previously visited position by comparing Scan Contexts at different time points. Global pose optimization based on graph optimization optimizes the global pose by constructing a pose graph and taking odometry, loop closure detection, and RTK data as constraints. Finally, the positioning accuracy and operational reliability of the proposed algorithm are verified through the positioning and mapping experiment of unmanned forklift-type mobile robots.