Remembering interaction episodes: An unsupervised learning approach for a humanoid robot

In this paper we will present a new approach to give a robot the capability to recognize already seen people and to remember details about past interactions. These details are time, length, location(GPS) and involved people of one interaction. Furthermore all features of this system work unsupervised. This means that the robot itself decides e.g. when and which person is important to remember or when an interaction starts. Out of these collected data additional information can be learned. For example a social network is build up which contains how often different people were seen together in the same interaction.