Evaluation of Simultaneous Localization and Mapping (SLAM) Algorithms
Garigipati, Chinna Bharath (2021)
2021
Automaatiotekniikan DI-ohjelma - Master's Programme in Automation Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2021-12-15
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202112089019
https://urn.fi/URN:NBN:fi:tuni-202112089019
Tiivistelmä
Simultaneous Localization and Mapping (SLAM) is a core component for the successful implementation of most autonomous robotics systems such as unmanned ground and areal vehicles and industrial mobile robots. It is crucial to have a robust localization and mapping system if the robot’s navigation depends on it while navigating through an unknown environment. However, the performance of a SLAM system varies based on the suitability of the algorithms to factors like environmental conditions, types of available sensors availability, dynamic motion, on-board computational capability of the robot, and more.
This thesis goes through the typical structure and components of a SLAM system and the factors affecting the performance. Various popular state-of-the-art LIDAR and visual SLAM algorithms and odometry algorithms that do not include loop closure are studied. The algorithm performance is tested using the datasets experiments conducted during the thesis. The custom datasets are collected using a multi-sensor setup include LIDARs, Stereo cameras, GNSS, and IMU sensors. Various experiments were conducted to simulate a few real-world conditions that a typical SLAM system could face, like rough terrains, different sensor elevations, varying robot velocities, and different mounting positions. Finally, the SLAM results on the custom data and the computational efficiency of the selected algorithms are analyzed using absolute position error and by recording the computational resource usage during runtime.
The purpose of this thesis is to provide an insight into the building blocks of the current state-of-the-art visual and LIDAR SLAM systems. Understand how external factors affect the system’s performance and how they compare with each other, and create a multi-sensor dataset to simultaneously test visual and LIDAR SLAM systems and that can be utilized for later research.
This thesis goes through the typical structure and components of a SLAM system and the factors affecting the performance. Various popular state-of-the-art LIDAR and visual SLAM algorithms and odometry algorithms that do not include loop closure are studied. The algorithm performance is tested using the datasets experiments conducted during the thesis. The custom datasets are collected using a multi-sensor setup include LIDARs, Stereo cameras, GNSS, and IMU sensors. Various experiments were conducted to simulate a few real-world conditions that a typical SLAM system could face, like rough terrains, different sensor elevations, varying robot velocities, and different mounting positions. Finally, the SLAM results on the custom data and the computational efficiency of the selected algorithms are analyzed using absolute position error and by recording the computational resource usage during runtime.
The purpose of this thesis is to provide an insight into the building blocks of the current state-of-the-art visual and LIDAR SLAM systems. Understand how external factors affect the system’s performance and how they compare with each other, and create a multi-sensor dataset to simultaneously test visual and LIDAR SLAM systems and that can be utilized for later research.