Active memory-based interaction strategies for learning-enabling behaviors

Despite increasing efforts in the field of social robotics and interactive systems integrated and fully autonomous robots which are capable of learning from interaction with inexperienced and non-expert users are still a rarity. However, in order to tackle the challenge of learning by interaction robots need to be equipped with a set of basic behaviors and abilities which have to be coupled and combined in a flexible manner. This paper presents how a recently proposed information-driven integration concept termed ldquoactive memoryrdquo is adopted to realize learning-enabling behaviors for a domestic robot. These behaviors enable it to (i) learn about its environment, (ii) interact with several humans simultaneously, and (iii) couple learning and interaction tightly. The basic interaction strategies on the basis of information exchange through the active memory are presented. A brief discussion of results obtained from live user trials with inexperienced users in a home tour scenario underpin the relevance and appropriateness of the described concepts.