Semantic Perception using Spatial Potential Fields

Dealing with knowledge from the human natural environment is one of the most important capabilities that robotic systems should be equipped with to act autonomously. Moreover, a robot cannot perform a given task properly without knowing the meaning of the contained objects. In addition, semantic perception is very challenging and robots must be able to deal with uncertainties and ambiguities that may occur partially caused by noisy sensors. In this paper the concept of a novel approach for semantic object recognition for indoor scenes is presented. In our method, spatial potential fields are defined to model probabilistic spatial relations between objects. These relations can then be used to recognise specific objects and find their most probable position in a given environment.