Autonomous exploration and mapping of unknown environments with teams of mobile robots

Resumo

This thesis covers the topic of autonomous exploration of unknown environments using mobile robots. It exposes the state of the art on this subject and classifies the techniques in terms of cooperation and integration with the SLAM algorithm. A comparison of the multiple techniques was performed with a specifically implemented software application. The results indicate the most appropriate techniques for different requirements. In addition, the thesis presents two new multi-robot coordinated and integrated exploration techniques. The first technique is, as far as we know, the first multi-robot behaviour-based reactive exploration method that is integrated with SLAM. The way to integrate the SLAM is by means of returning to previously explored zones when the uncertainty in the localization is high. The robot saves its pose when the localization is precise and, thus, it is able to return to these poses in order to reduce the uncertainty. However, we observe a loss in the exploration time efficiency due to the appearance of local minima in the potential field associated to the behaviour-based model. Furthermore, the technique is not fully scalable to explore large areas, since the full map has to be processed in real time. In this sense, a second technique that overcomes these problems was developed. It consists in a hybrid reactive/deliberative architecture in which the behaviour-based reactive control is limited to a local area. This new design avoids local minima and, since it does not use a full map in the reactive control, it is completely scalable. A deliberative planner is in charge of the long term planning. To this end, the map is divided using a tree where nodes represent positions with an associated area. This tree is used in order to decide if the robot should perform a local reactive exploration or move towards a long term goal. Finally, the thesis presents three application cases. The first case studies the problem of an automated search in an unknown environment using a collaborative control scheme in which the robots receive commands from an operator. In this way, the autonomous exploration planned by the robot also considers the urgency to explore the area indicated by the operator. In this way, it takes into account other factors, e.g., the localization uncertainty. The second application case studies the problem of searching dynamic agents. In this sense, it is developed a grid Bayesian filter that creates a map with the probability to find a target at each cell of the map. An adaptation of the deliberative planner used in the hybrid exploration model is used here in order to plan paths with this new map, thus leading the robot to the areas where it is likely to find targets. The third application case studies the problem of navigation in urban environments. Specifically, it is focussed in the problem of safely navigating in a pavement. We provide a solution to avoid falling in the curbstone and a pavement navigation strategy consisting in a directed exploration that plan straight paths in the pavement avoiding in this way the dropped kerbs, and garages or shop entrances.