Autonomous navigation of indoor wheeled robots based on improved Gmapping and improved Bidirectional A*

Abstract

Simultaneous Localization and Mapping (SLAM) and path planning are core technologies for autonomous mobile robots. Despite significant progress, low map accuracy and inefficient path planning remain challenges. The traditional Gmapping algorithm relies on odometry data for pose estimation, making it vulnerable to errors and leading to poor map accuracy. To address this, we propose the Multi-Sensor Fusion Improved Gmapping Algorithm (MSFI-Gmapping), which uses an Extended Kalman Filter (EKF) to fuse IMU angle data with odometry, improving pose estimation and enhancing map accuracy. Additionally, to tackle the low path search efficiency of the traditional A* algorithm, we introduce the Improved Bidirectional A* (IB-A*) algorithm. This algorithm dynamically combines Euclidean and Manhattan distances, enhances the heuristic function, adjusts weights based on obstacle density, and improves neighborhood search methods, thereby reducing redundant search nodes and improving path planning efficiency. Experimental results show that MSFI-Gmapping achieves a maximum average mapping error of 1.4% in a simulated environment and 1.9% in a real-world environment. Compared to the traditional A* algorithm, IB-A* reduces path planning time by approximately 87.5% and search node count by 46.2% in the simulation; in the real world, path planning time is reduced by 85.0% and search nodes by 48.1%.