A High-precision and High-robust Lidar-inertial SLAM Method Suitable for Robot Operation Scenarios

Abstract This paper proposes a tightly-coupled lidar-inertial odometry simultaneous localization and mapping (SLAM) framework, aiming to achieve high-precision and high-robustness position estimation and high-quality map construction in robot operation scenarios. In the feature extraction process, we innovatively adopt a face-based sampling strategy combined with uniform point cloud filtering to effectively optimize the spatial distribution characteristics of the point cloud data. When using the lidar odometry to correct the inertial measurement unit (IMU) data, we designed a lidar odometry confidence model based on the degree of degradation assessment, realized the optimization of the sensor fusion parameters through the dynamic noise covariance adjustment mechanism, and corrected the system state in real time by re-propagating the inertial data. In the optimization stage, we modify the residual computation model to appropriately increase the corner point matching weight and suppress the far-point noise. The proposed method is extensively evaluated and tested on the M2DGR dataset covering indoor and outdoor multi-scenes. Our scheme demonstrates advantages in terms of robustness and long-term error accumulation.