Full-day Workshop hosted at IEEE-Ro-MAN 2020, Naples, Italy
Endowing robots with learning and online behavioral adaptation abilities is a key objective for enabling natural and efficient human-robot interaction, especially in the areas of assistive and rehabilitation robotics. Nevertheless, as one of the novels of Asimov pointed out in The Complete Robot (1982), enabling a robot such as Lenny to learn inevitably leads to it making mistakes before adapting. One of the critical issues is thus to design robot learning and adaptation abilities that lead to a behavior that meets three criteria at the same time: efficiency, acceptability and security for the human.
To achieve efficiency, it is necessary to enhance the ability of the robots to adapt on the fly to the users’ behavior, needs and motivations. Social assistive robots have to incrementally learn user preferences and categories of user preferences experienced during past interactions (so that they do not start each new interaction from scratch), and to accordingly modify and adapt their behavior. This adaptation requires learning a model of human behavior and integrating this model into the decision-making algorithm of the robot. Efficient on the fly adaptation also requires model-free learning abilities to cope with local uncertainties of the environment, variations of the human’s desires and motivations, and volatilities of the interaction itself.
To achieve acceptability, robots have to be designed under a human-centered approach, which accounts for all the aspects that can affect the acceptability from the users’ point of view: from the users' expectation, to robot behavior legibility, to robot safe interactive behavior. Additionally, in the domain of assistive and social robotics and when working with vulnerable user populations, additional aspects have to be taken into account, such as ethical considerations and psychological/cognitive aspects. Both efficiency and acceptability can be enhanced if we consider the human-robot communication as a two-way street: both the user and the robot should be able to understand messages from the conversation partner, including the non-verbal ones. Meaning that the robot should be not only able to understand human activities, intentions and internal states, but also to show a similar behavior to humans.
Last but not the least, the security aspect have not to be neglected , since despite the recognized potential and usefulness of social robots for assistive purposes, people are still worried about their utilization also for fear of being victims of cybersecurity attacks, from privacy violation to remote robot tampering that can in the worst cases cause bodily harm.
The BAILAR workshop will constitute a unique opportunity to gather roboticists and computer scientists to present a variety of current approaches aiming at endowing social robots with learning, enhanced cognitive and social abilities, and discuss their potential to meet these criteria. This will permit to analyze the current state of the field and of its possible real-world applications in assistive and rehabilitation robotic.