Implementation of a Robust 3D LiDAR-Inertial Odometry and Mapping Algorithm: A ROS1 to ROS2 Conversion of FAST-LIO2
Open Access
- Author:
- Dougherty, Ryan
- Area of Honors:
- Electrical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Donald E Ebeigbe, Thesis Supervisor
Julio Urbina, Thesis Honors Advisor - Keywords:
- LiDAR
IMU
Mapping
3D Mapping
ROS
Robots
Odometry
FASTLIO
Robotics
Kalman Filter - Abstract:
- The ability for a robot to understand its surroundings is becoming more important in today’s world. LiDARs are sensors that can create a dense point cloud of the surrounding environment. On a mobile robot, a LiDAR itself is not sufficient to create an accurate map and localize itself inside of that map. Fusing LiDAR data with IMU measurements using a Kalman filter allows for extremely accurate mapping and odometry. FAST-LIO, a robust 3D LiDAR-Inertial Odometry and Mapping algorithm which is implemented using the Robot Operating System 1 (ROS1), is unable to be implemented on recent state-of-the-art robots which are mainly compatible with ROS2 (a more stable and efficient version of ROS1). This is also made worse by the fact that ROS1 is reaching its end of life and will no longer be supported in 2025. This necessitates the need for FAST-LIO to be rewritten using the ROS2 framework. This thesis converts FAST-LIO from the ROS1 framework to the ROS2 framework. The converted ROS2 package was tested using NVIDIA’s Isaac Sim, a photo realistic and physically accurate simulation software for robotics, as well as using prerecorded LiDAR and IMU data from a multi-scenario data set for ground robots. The accuracy of the mapping is visually tested using the results from both the data set and Isaac Sim. By comparing the trajectory from the new FAST-LIO to the ground truth provided by the data set, I verified the accuracy of the odometry and showed that the ROS2 version has a comparable accuracy to the ROS1 version.