| Ricardo Vazquez-Martin (rvmartin@uma.es) |
| Electronic Technology, University of Malaga |
| October, 2009 |
| Full text (external site) |
| He was born in Madrid, Spain, in 1975. He received the M.S. degree in Electronic Engineering and the PhD degree from the University of Malaga, Malaga, Spain, in 2002 and 2009, respectively. After some years working in companies related to industrial automation, in 2003 he returned to the University of Malaga to work as research assistant in the Electronic Technology Department. His research interests include visual SLAM, perception, mapping and mobile robot navigation. |
Abstract |
| This Thesis addresses different issues in the field of autonomous navigation. The achievement of this fundamental capacity will require the integration of various different modules of which perception, and localization and mapping are essential components of any navigation system. The work applies a feature-based representation to describe the environment, presenting two feature extraction algorithms for laser range-finding and vision. With respect to the localization and mapping problem, the laser features extracted are used to implement a SLAM algorithm. The mapping process of most SLAM approaches relies on metric information, leading to certain limitations related to computational complexity and error propagation. In this Thesis, a new map partitioning approach is presented based on the structure of the environment. In order to interact with the environment, robots usually carry onboard sensors to collect information (e.g., sonar arrays, laser range finder, vision). Understanding this sensory information is critical for other high-level tasks, including localization and mapping. An algorithm is presented for feature extraction and characterization from laser scan data. It is based on a curvature-based method where the raw data is segmented into sets of range readings using an adaptive curvature estimation technique. The proposed approach includes a curvature estimator to characterize the scan contour in order to achieve an affine-invariant laser scan segmentation. In a second approach, an affine region detector based on a perceptual grouping algorithm is applied to extract visual features. The image is pre-segmented in order to obtain an initial set of coloured blobs, which are then grouped into clusters in order to simplify image partition. Both approaches were tested in different environments, using different laser sensors and cameras, and were compared to some state-of-the-art algorithms. When there is available no independent source of location information (such as GPS), relative observations of the environment must be used. The SLAM technique solves this situation when a vehicle is moving in an unknown environment. An EKF based SLAM algorithm was implemented using the extracted laser features. In order to overcome the computational problems and error accumulation when dealing with metric maps, a map partitioning algorithm is proposed based on appearance information. The resulting map partition is thus in accordance with the structure of the environment. Observations gathered during the mapping process are used to build an auxiliary graph where the main property is the locality of features. The proposed map partitioning algorithm employs spectral clustering to find balanced partitions in this auxiliary graph. The approach generates an online submap to find partitions in those areas whose environments share a minimum amount of information. Finally, this algorithm is applicable to any kind of feature or sensor. |
ISSN: 1888-0258