In order to have a robotic system able to effectively learn by imitation, and not merely reproduce the movements of a human teacher, the system should have the capabilities of deeply understanding the perceived actions to be imitated. This paper deals with the development of a cognitive framework for learning by imitation in which a rich conceptual representation of the observed actions is built. The proposed architecture has been tested on the robotic system composed of a PUMA 200 industrial manipulator and an anthropomorphic robotic hand. The system demonstrated the ability to classify and imitate a rich set of movement primitives acquired through the vision system for manipulative purposes.

The paper deals with an artificial-vision based system for simultaneous people tracking and posture recognition in the context of human-computer interaction. We adopt no particular assumptions on the movement of a person and on its appearance, making the system suitable to several real-world applications. The system can be roughly subdivided into two highly correlated phases: tracking and recognition. The tracking phase is concerned with establishing coherent relations of the same subject between frames. We adopted the condensation algorithm due to its robustness in highly cluttered environments. The recognition phase adopts a modified eigenspace technique in order to classify between several different human postures and permits to exploit a rich interaction between a human user and a robotic system by means of a set of natural gestures. Experimental results have demonstrated the robustness of our system in different working conditions