A Software Architecture for Object Perception and Semantic Representation

In the near future, robots are expected to exhibit advanced capabilities when interacting with humans. In order to purposely understand humans and frame their requests in the right context, one of the major requirement for robot design is to develop a knowledge representation structure able to provide sensory data with a proper semantic description. This paper describes a software architecture aimed at detecting geometrical properties of a scene using an RGB-D sensor, and then categorising the objects within to associate them with a proper semantic annotation. Preliminary experiments are reported using a Baxter robot endowed with a Kinect RGB-D sensor.