Objectives: We had two specific aims: (1) to examine whether a robot co-therapist enhances outcomes for children with ASD compared to standard approaches involving only a human therapist, and (2) to examine whether in-session improvements can be generalized to natural environments.
Methods: Participants were 19 individuals with ASD between the ages of 6-13 years with varying levels of cognitive and language ability. Diagnoses were confirmed using the ADOS, SCQ-Lifetime, and clinical judgment. Participants completed 12 ABA therapy sessions as well as a baseline and posttest session. In six of the 12 sessions, the participant received therapy with a human therapist and a robot co-therapist, and from a human therapist only in the other six sessions. Order of presentation (robot + therapist first or therapist only first) was counterbalanced within matched pairs. Each session consisted of 30 to 40 minutes of ABA therapy administered by a social skills coach trained by a licensed psychologist and board certified behavior analyst with extensive experience working with individuals with ASD. In this study, we used the NAO platform (Aldebaran Robotics) as our robot co-therapist and controlled it with the DOMER program (Villano et al., 2011), a graphical user interface that allows a person to wirelessly control a NAO during a session. Using DOMER, the robot was controlled remotely by an unseen person, who observed the session and selected appropriate social responses for the robot. Measurements included in-session behavioral tracking, psychophysiological responses of the child as measured by an Affectiva Q Sensor, natural environment behavioral tracking sheets completed by the parents at several time points, and baseline/posttest evaluations of behavior using a number of parent report measures.
Results: Overall, between baseline and posttest, participants showed a decrease in social skills deficits, t(18)=2.07, p<.05, as measured by the Autism Social Skills Profile, and a trend toward a decrease in repetitive behaviors, t(18)=1.72, p<.10, as measured by the Bodfish Repetitive Behaviors Scale – Revised. While in-session behavioral tracking and electrophysiological responses showed considerable variability in performance, participants showed a marked increase in the frequency of parent-recorded targeted behaviors in their natural environment between baseline and posttest, t(17)=2.79, p<.01. Many participants showed greater improvement in targeted behaviors during sessions involving the robot, although this difference did not reach significance.
Conclusions: Overall, we found improvements in targeted behaviors and these gains generalized to natural environments, although it is important to note that there was considerable variability in behavioral and electrophysiological responses to the robot. It will be important to understand the factors that might contribute to this variability, and also to determine the ecological validity of this approach.