Estimación de incertidumbres en la elaboración de mapas con sensores LiDAR 2D
- María del Carmen Benavides Cuéllar Director
- José Alberto Benítez Andrades Director
Defence university: Universidad de León
Fecha de defensa: 25 October 2024
- José Ángel Hermida Alonso Chair
- Fuensanta Medina Domínguez Secretary
- Juan Carlos Álvarez Álvarez Committee member
Type: Thesis
Abstract
Robotic systems, or any system with a certain level of intelligence that interacts with the environment, need to receive information from their surroundings. These are often unknown. Mobile elements need to produce accurate maps based on the data provided by their sensors, while simultaneously estimating their position within the environment. This process is known as Simultaneous Localization and Mapping (SLAM). The robustness of the generated maps, as well as the element position and its associated uncertainty, are conditioned by the quality of the data captured by the sensors. Light Detection and Ranging (LiDAR) optical remote sensing systems use a polar measurement of distance and angle to characterise the surrounding elements. Infrared reflected radiation is used for the distance measurement. High measurement accuracy, compared to other positioning systems, as well as the high data density provided, has turned LiDAR into a common alternative in robotic systems, autonomous vehicles or consumer electronic devices. These characteristics allow for the compilation of highly detailed maps and the estimation of position within them. However, this high level of detail comes at the expense of managing high volumes of information, which considerably increases the computational burden of the SLAM process. Two research trends can be identified in the processing of LiDAR data for SLAM purposes. Some researchers work directly with the entire raw dataset, seeking the best match between scans by evaluating the distance between point clouds and minimising it with mathematical optimisation techniques. Other authors simplify the sensor information to a small set of geometric features that uniquely define the mapped profile. The present doctoral thesis focuses on the mapping stage of the SLAM process, based on the information provided by two-dimensional LiDAR sensors. The first general objective of this work is to simplify the environment into an invariant polygonal set of straight sections and virtual intersections capable of defining it as efficiently and accurately as possible. For this purpose, we present two geometrical primitive extraction methods. The proposed methodologies are Weighted Conformal LiDAR-Mapping (WCLM) and Conditional Weighted Linear Fitting (CWLF). These are compared, in terms of computational and uncertainties, with two classical reference methods. WCLM methodology estimates the parameters that define environment straight sections and their intersections in the inverse complex domain. This method gives greater weighting to points close to the straight line than to outliers, following a bivariate distribution. The CWLF methodology obtains the same characteristic elements as the WCLM. In this case, the process is based on conditional linear regression, considering zero error of the independent variable. A conditional probability distribution is followed by CWLF, simplifying the methodology with respect to WCLM. The second general objective of this research is to evaluate the influence of the 2D LiDAR sensor scanning frequency during the feature extraction process. The quality of the detected significant points, as well as processing computational burden, are affected by the value of the scanning frequency. This parameter, variable in some commercial sensors, must be adjusted according to the application requirements. The analysis of the integration of Robot Operating System (ROS) with MATLAB for the teleoperation of a robotic system equipped with a camera and a 2D LiDAR sensor is the last general objective. The remote control of the system is performed using MATLAB. For this purpose, a human-machine interface has been developed to visualise the data acquired by the sensors, the data processed by ROS or MATLAB, as well as the implementation of a virtual joystick to control the robot remotely. Among the data processed in MATLAB, geometric features extracted with the WCLM methodology have been represented in real time on the compiled environment map.