Exploration and Mapping of Spatio-Temporal Pedestrian Flow Patterns for Mobile Robots

Socially compliant robot navigation is one of the key aspects for long-term acceptance of mobile robots in human-populated environments. One of the current barriers for this acceptance is that many navigation methods are based only on reactive behaviours, which can lead to frequent re-plannings, causing an erratic or aggressive robot behaviour. Instead, giving the ability to model and predict in advance how the people are likely to behave, from a long-term perspective, is an important enabler for safe and efficient navigation. For example, a robot may use its knowledge of the expected human motion to go with the main direction of flow to minimise the possibility of collisions or trajectory re-plannings. In order to provide robots with knowledge of the expected activity patterns of people at different places and times,the first main contribution of this thesis is the introduction of a Spatio-Temporal Flow map (STeF-map). This is a time-dependent probabilistic map able to model and predict the flow patterns of people in the environment. The proposed representation models the likelihood of motion directions on a grid-based map by a set of harmonic functions, which efficiently capture long-term variations of crowd movements over time. The experimental evaluation shows that the proposed model enables a better human motion prediction than spatial-only approaches and an increased capacity for socially compliant robot navigation. Obtaining this knowledge from a mobile robot platform is, however, not a trivial task, as usually they can only observe a fraction of the environment at a time, while the activity patterns of people may also change at different times. Therefore, the second main contribution is the investigation of a new methodology for mobile robot exploration to maximise the knowledge of human activity patterns, by deciding where and when to collect observations based on an exploration policy driven by the entropy levels in a spatio-temporal map. The evaluation is performed by simulating mobile robot exploration using real sensory data from three long-term pedestrian datasets, and the results show that for certain scenarios, the proposed exploration system can learn STeF-maps more quickly and better predict the flow patterns than uninformed strategies.