Resumo:
In recent years, the development of simultaneous localization and mapping (SLAM) algorithms using data from LiDAR (Light Detection Ranging) sensors has been increasing. LiDAR sensors are capable of performing 3D scanning in 360° of environments, with relatively low acquisition time and energy consumption, which allows their use in embedded systems. The objective of this research is to study and implement SLAM techniques using LiDAR sensor data, to obtain mapping and localization in indoor environments for navigation tasks with mobile robots. In this context, the LeGO-LOAM, A-LOAM and F-LOAM techniques were implemented, aiming to analyze the influence of the volume of static geometric references features on the quality of the mapping and accuracy of the instantaneous localization. The LOAM strategies capture geometric features from the classification of the points of the cloud captured by the LiDAR sensor, and subsequently identify correspondence between consecutive scans to obtain the location estimate. To analyze the performance of the techniques, they were tested in simulation and in a real environment. The simulator used was the Gazebo, the Husky mobile robot, the Velodyne VLP-16 LiDAR sensor, with scans performed in the following environments: a closed straight corridor with windows, a closed straight corridor without windows, a closed square corridor and a 3D model of LaR - Robotics Laboratory of DEEC/UFBA. The real experiments were conducted in LaR, using the Husky mobile robot, the Velodyne VLP-16 LiDAR sensor and the Optitrack 3D motion capture system. The experiments confirmed the efficiency of the LiDAR SLAM algorithms for localization and mapping of mobile robots, with better performance in environments with many static references. In simpler scenarios, the results were also satisfactory, except for the LeGO-LOAM algorithm, which presented greater error on the x-axis in straight corridors without windows due to the lack of references. The LeGO-LOAM algorithm stood out in the generated maps, displaying greater detail of the structures and objects present in the environment. Despite a tendency to present more significant errors on curves, the evaluated LiDAR SLAM algorithms demonstrated efficiency in correcting estimates on straight stretches, evidencing a low susceptibility to accumulated errors. These results reinforce the potential of these techniques as reliable solutions for accurate mapping and localization in complex environments.
