A Context-Aware Adaptability Model for Service Robots Gonc ¸alo S. Martins 1 , Paulo Ferreira 1 , Lu´ ıs Santos 1 , Jorge Dias 1,2 1 Institute of Systems and Robotics, University of Coimbra, 3030-290 Coimbra, Portugal 2 Khalifa University, Robotics Institute, Abu Dhabi, UAE {gmartins, paulo.ferreira, luis}@isr.uc.pt, jorge.dias@kustar.ac.ae Abstract This article presents a context-aware, adaptable ser- vice selection model for a social robot, giving it the ability to estimate the user’s expectation, as- sess the degree of satisfaction and use it as feed- back to improve subsequent interactions. We es- tablished specific measures for expectation and sat- isfaction, estimated using Bayesian inference and used to control the human-robot interaction. This work is proposed to overcome the fact that ser- vice robots are usually designed to perform within a very strict operational envelope, sometimes re- quiring all knowledge to be known and locally pre- programmed. The model was tested in demand- ing simulated scenarios, showing promising results, and also in exploratory experiments with users. 1 Introduction In the past few years, we have witnessed an increased inter- est from the research community on developing intelligent robotic systems, such as the Social Robots and Compan- ion Robots used in applications of Ambient Assisted Living. Such systems can be used as a technical solution to support our society and organizations to tackle the challenge posed by the continuously growing fraction of elderly population in developed countries. Several solutions have been devised and successfully tested. However, these solutions tend to depend greatly in previous knowledge and are unable to interact with the user in an adaptable way. In this work, we aim to contribute towards autonomy in interaction of a service-providing Social Robot. More specif- ically, we tackle the issue of autonomously selecting the ser- vice that should be performed to a user in order to ensure their satisfaction. We argue that satisfaction can be fulfilled when the action that is performed matches the user’s expecta- tion. These two concepts are used to perform a limited form of interaction regulation. This work was developed in the context of the GrowMeUp project, funded by the European Union’s Horizon 2020 Research and Innovation Programme - Societal Challenge 1 (DG CON- NECT/H) under grant agreement N o 643647. The remainder of this work is organized as follows. Sec- tion 2 presents a short review of recent work on User Adapt- ability. Section 3 presents the proposed model. Section 4 presents our experiments, including the particular instantia- tion of the model used in them and the results we have ob- tained. These results are then discussed in Section 5, fol- lowed by our concluding remarks and notes on future work, in Section 6. 2 Related Work In this section we present works involving robotic systems that are User Adaptable, a characteristic here defined as the system’s ability to automatically adapt its interaction to its users. There have been studies on how users can coexist with robots and on what characteristics these should exhibit in or- der for the user to accept them as a social entities [de Graaf et al., 2015][de Graaf and Ben Allouch, 2013][Fischinger et al., 2014]. These studies show that humans are indeed capable, and willing to, attribute a social role to robots and to perceive them as part of their social circle. More importantly for this work, [Heerink, 2011] shows that users of social robot sys- tems prefer system-controlled to user-controlled adaptation, although they do prefer to maintain a sense of control. Focusing on children as end-users, the authors of [Kanda et al., 2004] and [Kanda et al., 2007] present the results of long- term trials involving child students, during which the robots were able to adapt their behaviors using a “pseudo develop- ment” system, in accordance with the interactions that they experienced with the children. The authors of [Ros et al., 2014] and [Magyar and Vircikova, 2015] focus on solutions for robotic dance tutors for children, with the first focusing on learning the policy of a Wizard of Oz, and the second on implementing a system that adapts based on previous interac- tions and a user model. The authors of [Baraka and Veloso, 2015] also take a step towards long-term cohabitation of robots and humans, by pre- senting a system that aims at autonomously adapting to the user’s preferences over a long-term period where several in- teractions occur. In order to aid in the adaptation, the authors explicitly rely on the user providing a rating of the robot’s behavior, which is then used to learn the parameters of the model employed. A number of works make use of the user’s personal- ity [Eysenck, 1991] to achieve higher levels of adaptation,