Chen Jianhong, Wang Hongwei, Yang Shan
School of Resources and Safety Engineering, Central South University, Changsha 410083, China.
Sensors (Basel). 2023 Jul 31;23(15):6834. doi: 10.3390/s23156834.
The demand for autonomous exploration and mapping of underground environments has significantly increased in recent years. However, accurately localizing and mapping robots in subterranean settings presents notable challenges. This paper presents a tightly coupled LiDAR-Inertial odometry system that combines the NanoGICP point cloud registration method with IMU pre-integration using incremental smoothing and mapping. Specifically, a point cloud affected by dust particles is first filtered out and separated into ground and non-ground point clouds (for ground vehicles). To maintain accuracy in environments with spatial variations, an adaptive voxel filter is employed, which reduces computation time while preserving accuracy. The estimated motion derived from IMU pre-integration is utilized to correct point cloud distortion and provide an initial estimation for LiDAR odometry. Subsequently, a scan-to-map point cloud registration is executed using NanoGICP to obtain a more refined pose estimation. The resulting LiDAR odometry is then employed to estimate the bias of the IMU. We comprehensively evaluated our system on established subterranean datasets. These datasets were collected by two separate teams using different platforms during the DARPA Subterranean (SubT) Challenge. The experimental results demonstrate that our system achieved performance enhancements as high as 50-60% in terms of root mean square error (RMSE).
近年来，对地下环境自主探索和测绘的需求显著增加。然而，在地下环境中精确定位和测绘机器人面临着显著挑战。本文提出了一种紧密耦合的激光雷达-惯性里程计系统，该系统将NanoGICP点云配准方法与使用增量平滑和映射的惯性测量单元（IMU）预积分相结合。具体而言，首先滤除受灰尘颗粒影响的点云，并将其分离为地面和非地面点云（用于地面车辆）。为了在空间变化的环境中保持精度，采用了自适应体素滤波器，该滤波器在保持精度的同时减少了计算时间。利用从IMU预积分得出的估计运动来校正点云畸变，并为激光雷达里程计提供初始估计。随后，使用NanoGICP执行扫描到地图的点云配准，以获得更精确的位姿估计。然后，将所得的激光雷达里程计用于估计IMU的偏差。我们在已建立的地下数据集上对我们的系统进行了全面评估。这些数据集是由两个独立的团队在DARPA地下（SubT）挑战赛期间使用不同平台收集的。实验结果表明，我们的系统在均方根误差（RMSE）方面实现了高达50%-60%的性能提升。
