Dispelling Illusions of Self-Motion: Robust Localisation for Autonomous Robots in Feature-Sparse and Dynamic Indoor Environments

Abstract

Feature-sparse and dynamic environments pose significant challenges for autonomous robots. This thesis presents an online method for removing dynamic objects from 2D LiDAR measurements. The dynamic filter first segments and tracks objects in the data and uses a Kalman filter to estimate their speed and determine whether each object is static or dynamic. Testing shows that the dynamic filter improves localisation performance in a feature-sparse and dynamic environment. Additionally, a nonholonomic Coordinated Turn (CT) model was implemented in the robot_localization (RL) library for ROS 2 to improve odometry pose estimation. Testing showed no noticeable difference compared with RL’s default omnidirectional model.