Automatic Learning of Cognitive Exercises for Socially Assistive Robotics

Alejandro Suárez-Hernándeza , Antonio Andriellaa , Aleksandar Taranovićb, Javier Segovia-Aguasc, Carme Torrasa, and Guillem Alenyàa

a Institut de Robòtica i Informàtica Industrial, CSIC-UPC, C/ Llorens i Artigas 4-6, 08028 Barcelona, Spain

b Bosch Center for Artificial Intelligence, Renningen, Germany

c Universitat Pompeu Fabra, Tànger 122-140, 08018 Barcelona, Spain

Abstract: In this paper, we present a learning approach to facilitate the teaching of new board exercises to assistive robotic systems. We formulate the problem as the learning of action models using Boolean predicates, disjunctive preconditions, and existential quantifiers from demonstrations of successful exercise executions. To be able to cope with exercises whose rules depend on a set of features that are initialized at the beginning of each play-out, we introduce the concept of \textit{dynamic context}. Furthermore, we show how the learnt knowledge can be represented intuitively in a graphical interface that helps the caregiver understand what the system has learnt. As validation, we conducted a user study in which we evaluated whether and to which extent different types of feedback can affect the subjects' performance while teaching three types of exercises: (1) sorting numbers; (2) arranging letters; and (3) reproducing shapes sequences in reversed order. The results suggest that textual and graphical feedback are beneficial.


Very commonly we find that robotics systems are hard to deploy in new applicactions without encoding the necessary domain control knowledge as a program. This obstacle is very prominent in Socially Assistive Robotics (SAR), a field whose main goal is assisting users through social interaction (e.g. encouraging, instructing, and monitoring) rather than physical one. Effective SAR applications often require the ability to learn new behaviors tailored for certain users. In this work we present INPRO (Intuitive Programming), an algorithm that allows a robot to learn the rules of an exercise executed on a board. These exercises are reminiscent of the Syndrom Kurztest (SKT), and are meant to track the cognitive ability of patients. Ultimately, several socially assistive robots can help a caregiver administer these tests to several patients at once. However, a caregiver may wish to tailor some tests to address the specific needs of a group of patients. Thus, the target audience of our framework is caregivers who are in this need to extend the repertoire of exercises that the robot can administer. Our method takes as input several examples demonstrated by the caregiver of how an exercise is carried out. These inputs are processed using modern infers from them the underlying rules that govern the exercise. Thus, INPRO belongs to the category of algorithms that allows a robot to be programmed without technical expertise.


It will be available soon.

Source code

It will be available soon.