Scan Matching Confidence Evaluation for Robust LiDAR Odometry and Pose Graph Optimization

In this study, we propose a method for scan matching confidence evaluation in LiDAR SLAM systems and apply it to both odometry and pose graph optimization to improve overall trajectory accuracy. Conventional LiDAR SLAM systems apply a fixed weight to LiDAR measurements regardless of the scan matching quality, which may lead to drift errors or even divergence in environments with narrow spaces or limited fields of view. Thus, we introduce the concept of dilution of precision—widely used in GNSS—to evaluate scan matching reliability based on the geometric distribution between matched points and the LiDAR sensor. The quantified reliability is used to adjust the measurement weight in Kalman filter-based odometry and to filter out unreliable factors during loop closure, preventing erroneous relative poses from being incorporated into the pose graph. The proposed method was evaluated on the Hilti SLAM Challenge 2022 and VBR SLAM Dataset. Additionally, it improves performance regarding drift suppression and trajectory accuracy based on absolute trajectory error analysis.