Research on Improved 2D SLAM Algorithm based on Gmapping

Abstract

In the traditional Gmapping algorithm, particle filtering is susceptible to motion noise, and particles may degrade or fall into local optima, leading to decreased mapping accuracy. This paper proposes an improved Gmapping method that integrates Extended Kalman Filter (EKF) prediction with Adaptive Particle Swarm Optimization (APSO) to enhance SLAM accuracy and robustness. First, EKF is used to fuse IMU and wheeled odometry information for pose prediction, ensuring more precise particle initialization and reducing the impact of motion noise on particle sampling. Second, APSO dynamically adjusts particle velocity and pose to improve global search capability, preventing premature convergence. Finally, APSO is introduced in the resampling process to optimize particle diversity and mitigate particle degradation. Experimental results demonstrate that the proposed algorithm achieves higher localization accuracy and mapping quality in long corridors and low-feature areas while reducing computational cost and enhancing SLAM system robustness.