Our goal is to develop the fundamental approaches required to design mobile robot systems that can reliably operate over extended periods of time in dynamically changing environments. To achieve this, robots need the ability to learn and update appropriate models of their environment including the dynamic aspects and to effectively incorporate all the information into their decision-making processes. The time is ripe for the next step in navigation: the algorithms for state estimation and navigation in static environments have reached a high level of sophistication and the underlying models and learning algorithms are well-understood. Our goal is to take these approaches further and to develop effective and object-oriented three-dimensional representations, that cover all aspects of the dynamic environment required for reliable and long-term mobile robot navigation. The outcome of this research will be relevant for all applications that are based on autonomous navigation in real-world scenarios including autonomous robots, mobile manipulation, transportation systems, or autonomous cars.