Identifying engagement from joint kinematics data for robot therapy prompt interventions for children with autism spectrum disorder

Abstract

Prompts are used by therapists to help children with autism spectrum disorder learn and acquire desirable skills and behaviors. As social robots are more regularly translated into similar therapy settings, a critical part of ensuring effectiveness of these robot therapy system is providing them with the ability to detect engagement/disengagement states of the child in order to provide prompts at the right time. In this paper, we examine the various features related to body movement that can be utilized to define engagement levels and develop a model using these features for identifying engagement/disengagement states. The model was validated in a pilot study with child participants. Results show that our engagement model can achieve a recognition rate of 97 %.