Design of an autobiographical memory for a social robot supporting human-robot interactions

Abstract

Arti cial companions harnessed with long-term interaction capabilities are useful for a variety of applications. The ability of recalling past memories during ongoing interac- tions and adapting behavior according to the interaction partner are the reinforcements of a successful long-term interaction. Memory has been gured out as the underlying mechanism which governs these behaviors. Even though a number of e orts have been taken, the capabilities of existing arti cial companions have not reached to human level. Modeling the memory is still remains as one of the challenges for achieving long-term human robot interaction (HRI). Memory based system have been designed for remem- bering users, their preferences and past emotionally salient events with them. However, these systems face di culties when interacting with a group of users. They have certain limitations including remembering user groups, relationships between users to mention a few. The requirement of memory model that has human-like capabilities has not been ful lled yet. This work presents an Autobiographical Memory (AM) based intelligent system which can be applied for HRI. The AM comprises of three layers namely self layer, people layer and episode layer. Methods have been developed for extracting, storing, updating and recalling user information during HRI. A system has been designed for learning user preferences through human friendly interactions and providing user adaptive services for each user in a multi-user domestic environment. Furthermore, the system is capable of adapting according to users hidden preference and changes of preferences. The robots memory has been structured in such a way that it can easily remember the user groups and the relationship between users. The proposed AM is also capable of remembering spatial information and sequence of past actions. A novel method has been proposed for arranging a set of objects in a surface while interpreting uncertain spatial and qualitative distance information in user commands. Performance of the system has been validated by using a set of experiments. The proposed AM based intelligent system is capable of supporting long-term human- robot interactions.