Social situation-aware perception and action for cognitive robots

This deliverable describes the work on task T2.4 done after what was reported in deliverable D2.6. We have mainly pursued two directions of work during 2015. The first direction of work was research on socially annotated mapping, with focus on human motion patterns. We have developed a method to model velocity patterns in the environment using Gaussian mixture models and worked on extending it with usage of Markov Chains. The Gaussian mixture model models the dominant motion patterns (direction and speed) while the associated Markov Chain represents the temporal relation between directions of motion (“exit velocity, conditioned on entry velocity”). We have also focused on improving localization robustness and optimizing mapping and localization modules. A major extension has been to employ a 3D sensor for mapping and localization purposes, in order to overcome problems with extreme occlusion of the 2D sensors at leg height during busy periods in the airport.