Mitigating Perceptual Drift in ROS2 LiDAR-Based SLAM During Wall-Parallel Motion

Autonomous mobile robots that rely on 2D LiDAR-based SLAM frequently exhibit instability when navigating long planar corridors or moving parallel to walls. This phenomenon, commonly observed in ROS2-based TurtleBot platforms, leads to erroneous localization updates, lateral drift, and random heading shifts, collectively termed perceptual drift or hallucination. This paper investigates the underlying causes of such errors in the AMCL and Cartographer frameworks, identifying key contributors including geometric degeneracy, scan-matching ambiguity, and improper covariance synchronization between odometry and LiDAR models. A reproducible simulation benchmark was developed in Gazebo using TurtleBot3 and Nav2 to systematically evaluate these behaviors. Experimental results demonstrate that fusing IMU and wheel-odometry data through robot_localization, combined with adaptive AMCL parameter tuning and LiDAR scan pre-filtering, significantly mitigates drift while maintaining computational efficiency. Quantitative metrics show reductions in localization RMSE by up to 37% and recovery-rate improvements exceeding 40% compared to default configurations. The findings provide actionable insights for enhancing SLAM robustness in constrained planar environments.