Large scale localization and mapping for deployment of robotic wheelchair in a building

• Main Author: Lim, Jit Ern
• Other Authors: Ang Wei Tech
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; SLAM; Localization; Mapping; Sensors; Mobile robot; Robotics; ROS
• Online Access: https://hdl.handle.net/10356/176826
• Institution: Nanyang Technological University

Gmapping and AMCL is a popular SLAM and localization combination in ROS1 that is used for the deployment of autonomous mobile robots. However, Gmapping is not suited for large environments as the mapping process has to be restarted from scratch and completed in one run. Furthermore, the mapping process has to be repeated once they are significant changes in the environment to maintain the localization accuracy of AMCL. Therefore, researchers have developed multisession and lifelong SLAM approaches for real-world operations in large environments. However for the deployment of a robotic wheelchair, where cost is also an important criteria along with safety, it is not clear which multisession and lifelong SLAM and localization method and sensor configuration is most suitable. In this paper, state-ofthe-art SLAM with lifelong and multisession functions are reviewed through literature to select a suitable approach. For our specific application, RTAB-Map is selected because it can receive inputs from multiple sensor combinations. It is subsequently used to determine the sensors required by performing mapping and localization in two large environments that are indoor and semi-outdoor. The hardware utilized is a robotic wheelchair equipped with these sensors: 2D LiDAR, 3D LiDAR, three RGB-D cameras, wheel encoders and IMU. For mapping, the quality of 2D occupancy grids is evaluated based on the geometry of the map and the representation of obstacles in the map to reduce the amount of manual postprocessing required. The sensor combination that created the best quality map is then changed to cheaper sensor subsets and compared among each other during localization to generate a guideline for sensor selection based on cost constraints and localization performance requirements. Then, the fused wheel encoder and IMU odometry is determine to be necessary and cannot be removed to further reduce cost after comparing their performance with exteroceptive odometry. Finally, the deployment of robotic wheelchair in a large environment is made more feasible due to the reduced cost using RTAB-Map with 3D LiDAR and two RGB-D cameras for mapping and RTAB-Map with 2D LiDAR and one RGB-D camera for localization with fused wheel encoder and IMU odometry for both cases.