Towards Long-term Autonomy through Temporal modeling

In our project, we will develop methods that can introduce the notion of time into discrete and continuous representations used in the mobile robotics domain. Thus, representations will not only describe the environment structure but also how it changes over time. Knowledge of the dynamics will allow performing long-term predictions of the environment states and their uncertainties, which will improve the robustness of localization and efficiency of planning. In our project, we will combine the expertise of the IRAP@KAIST team in reliable localisation and large-scale mapping, with the principles of frequency-based spatio-temporal modeling, developed by the CTU Team The combination of the aforementioned methods will result in models which can efficiently represent the evolution of the world over time, as observed through the robotic sensors, and predict the future structure of the environment. Unlike previous efforts that aimed to model the changes in the environment’s appearance though representation- specific methods, we will focus on models of changes in the environment structure.