Exploration and spatial learning in dynamic environments

Mobile robots have the potential to be useful in a wide variety of domains, from delivering packages in office buildings to driving vehicles on highways to performing reconnaissance on battlefields. If mobile robots are to live up to their potential, they need to be able to deal with the changes that may occur in these environments, whether those changes consist of people walking about or bridges being washed away. This dissertation presents techniques developed for exploration, learning, and navigation in dynamic environments. Adaptive place networks, topological/metric maps incorporating variable-confidence adjacency links, are introduced as a means for dealing with topological changes. ELDEN is an integrated system that combines adaptive place networks with a reactive, behavior-based controller for dealing with transient changes and a relocalization system for correcting dead reckoning error. ELDEN has been implemented on a real mobile robot and has been able to explore and navigate successfully in a dynamic, real-world environment. Strategies for directing exploration are also presented, along with quantitative simulation experiments measuring the performance of these strategies.