Context-Dependent Social Mapping

Advanced robotics systems in human frequented environments need to be equipped with avant-garde capabilities, so as to attain a social behavior acceptable by their human proprietors. Therefore, robots should learn and react properly when they share a common domain with humans and adjust their operation according to the activity of the people around. This paper proposes a social mapping method which makes use of 3D maps, action recognition and proxemics theory. In particular, as it moves around, the robot builds a metric map of the surroundings and arranges it in topological graphs. Meanwhile, it can detect any individual existing nearby and capitalizes on a deep learning technique to recognize its action. The recognized actions form the context-dependent social zones which are registered with specific proxemics rules to determine the robot's navigational behavior. The proposed method was assessed with an indoors navigating robot, equipped with an RGB-D sensor. The human and action recognition units showed superior operation rendering the navigation module capable of planning trajectories for the robot to move around humans.