Building Visual Maps with a Team of Mobile Robots

摘要

This paper tackles the problem of Simultaneous Localization and Map Building (SLAM) carried out by a team of robots. Particularly these robots build landmark-based maps by extracting interest points from the environment. These points are characterized by a local descriptor and a 3D position on the environment, constituting the visual landmarks. In this approach we consider the situation in which the robots start their mapping task independently. That is to say, the path followed by the robots and the observations are estimated independently. After a while, there is a set of independent local maps that can be fused in order to obtain a global map. For that reason, we have also solved the problem of aligning and fusing visual maps. Finally, once a global map is obtained, the robots continue the map building task jointly. Regarding the sensors used in order to extract information from the environment, some authors use range sensors such as SONAR (Wijk & Christensen, 2000; Kwak et al., 2008) or LASER (Leonard & Durrant-Whyte, 1991; Thrun, 2001). However, in the last years, there is a great interest on using cameras as sensors. This approach is denoted as visual SLAM (Valls-Miro et al., 2006). The cameras are less expensive than laser and offer a higher amount of information from the environment. This advantage makes it possible to incorporate additional applications to the robot, such as face recognititon. Additionally, 3D information can be obtained from the environment when using stereo vision (Murray & Little, 2000; Gil, Reinoso, Fernandez, Vicente, Rottmann M Little et al., 2002) and SURF (Murillo et al., 2007). In (Gil et al., 2009; Ballesta, Gil, Reinoso & Ubeda, 2010) we performed an evaluation comparing several detection and description methods in order to obtain the most suitable feature extractor for visual SLAM. As a result, we obtained that the Harris Corner Detector in combination with the u-SURF descriptor satisfied our requirements in visual SLAM. The task of building a map of the environment while simultaneously localizing in it can be performed by a single robot. However, it will be carried out with more efficiency if there is a team of robots which collaborates in this task. This approach is denoted as multi-robot SLAM (Konolige et al., 2003). In this field, two main solutions can be found. On the one hand, in some proposals the robots build a unique global map (Gil, Reinoso, Martinez-Mozos, Stachniss & Burgard, 2006; Fenwick et al., 2002). In this case, the exploration task can be performed more efficiently since the robots have a global notion 6