Dynamic Residual Calibration and Multi-Scale Fusion for Accurate Prediction of Non-Stationary Backend Indicators

Abstract

This study presents an enhanced Cartographer algorithm for simultaneous localization and mapping (SLAM) applied to multi-sensor robotic navigation systems. The proposed approach integrates LiDAR, odometer, and inertial measurement unit (IMU) data within a unified mapping framework to improve localization accuracy and environmental perception. To mitigate common issues such as occlusion, reflection, and measurement noise in LiDAR point clouds, the Point Cloud Library (PCL) is utilized for preprocessing to remove outliers and dropout points. An adaptive filtering mechanism based on the dynamic window method enables real-time obstacle avoidance and trajectory correction during autonomous navigation. The system is simulated under the ROS framework using Rviz and Gazebo environments, where results demonstrate superior mapping clarity and reduced drift compared with the original Cartographer algorithm. Field experiments confirm that the optimized approach effectively eliminates abnormal laser points, enhances robustness, and produces high-precision maps suitable for complex dynamic environments. Overall, this work provides a practical and extensible solution for intelligent robotic navigation and SLAM optimization in multi-source perception scenarios.