|
main page |
AbstractModels of the environment are needed for a wide range of robotic applications, from search and rescue to automated vacuum cleaning. Learning maps has therefore been a major research focus in the robotics community over the last decades. Robots that are able to acquire an accurate model of their environment on their own are regarded as fulfilling a major precondition of truly autonomous agents. In order to solve the map learning problem, a robot has to address mapping, localization, and path planning at the same time. In general, these three tasks cannot be decoupled and solved independently and map learning is thus referred to as the simultaneous planning, localization, and mapping problem. Because of the coupling between these tasks, this problem is very complex. It can become even more complex when there are dynamic changes in the environment or several robots are being used together to solve the problem.The contributions of this thesis are solutions to various aspects of the autonomous map learning problem. We first present approaches to exploration that take into account the uncertainty in the world model of the robot. We then describe how to achieve good collaboration among a team of robots so that they efficiently solve an exploration task. Our approach distributes the robots over the environment and in this way avoids redundant work and reduces the risk of interference between the individual team members. We furthermore provide a technique to make use of background knowledge about typical spacial structures when distributing the robots over the environment. As a result, the overall time needed to complete the exploration mission is reduced. To deal with the uncertainty in the pose of a robot, we present a solution to the simultaneous localization and mapping problem. The difficulty in this context is to build up a map while at the same time localizing the robot in this map. Our approach maintains a joint posterior about the trajectory of the robot and the model of the environment. It produces highly accurate maps in an efficient and robust way. In this thesis, we address step-by-step the different problems in the context of map learning and integrate our techniques into a single system. We provide an integrated approach that simultaneously deals with mapping, localization, and path planning. It seeks to minimize the uncertainty in the map and in the trajectory estimate based on the expected information gain of future actions. It takes into account potential observation sequences to estimate the uncertainty reduction in the world model when carrying out a specific action. Additionally, we focus on mapping and localization in non-static environments. Our approach allows a robot to consider different spatial configurations of the environment and in this way makes the pose estimate more robust and accurate in non-static worlds. |