Interactive robots with model-based 'autism-like' behaviors: Assessing validity and potential benefits

Abstract

Due to their predictability, controllability, and simple social abilities, robots are starting to be used in diverse ways to assist individuals with Autism Spectrum Disorder (ASD). In this work, we investigate an alternative and novel research direction for using robots in relation to ASD, through programming a humanoid robot to exhibit behaviors similar to those observed in children with ASD. We designed 16 'autism-like' behaviors of different severities on a NAO robot, based on ADOS-2, the gold standard for ASD diagnosis. Our behaviors span four dimensions, verbal and non-verbal, and correspond to a spectrum of typical ASD responses to 3 different stimulus families inspired by standard diagnostic tasks. We integrated these behaviors in an autonomous agent running on the robot, with which humans can continuously interact through predefined stimuli. Through user-controllable features, we allow for 256 unique customizations of the robot's behavioral profile.We evaluated the validity of our interactive robot both in video-based and 'in situ' studies with 3 therapists. We also present subjective evaluations on the potential benefits of such robots to complement existing therapist training, as well as to enable novel tasks for ASD therapy.