Robot-Mediated Adaptive Response System in Joint Attention Task for Children with Autism Spectrum Disorders

Background: Best practices in autism treatment call for intensive and individualized interventions. The human resources (20-25 hours) required to achieve this intervention, however, are expensive. Further, the number of children who need intensive services is increasing. The field of robotic science may expand the capabilities of humans in several ways. Research shows that for children with ASDs, human therapists may evoke less frequent and shorter durations of attention than do robots. Incorporating robots in intervention may lead to an increase in engagement making intervention more efficacious. Also, robots may be capable of micro level behavior discrimination beyond that of human capacity, making robot-assisted intervention more efficient.

Objectives: To determine the feasibility and potential value of humanoid robotic intervention system we developed a co-robotic system capable of administering joint attention tasks to young children with ASD.

Methods: We developed a test-bed that consisted of a humanoid robot NAO, 4 Infrared (IR) cameras mounted on the wall and ceiling, and networked computer monitors that displayed task stimuli. A baseball hat with arrays of IR LEDS was employed to determine the child’s gaze. We compared performance and gaze detection for a sample of 6 children with clinically confirmed ASD diagnosis and 6 typically developing (TD) children (all ages 3-5). A series of joint attention prompts were administered via either a human (x2) or the humanoid robot (x2) with randomized presentation. The child sat in a chair across from the robot or interventionist and was instructed through a hierarchy of prompts (i.e., gaze shifts, pointing prompts, prompts with target activation) to look to a target. 

Results: The system registered gaze across all trials, looks to target (monitor on wall), and provided reinforcement for target responses (i.e., verbal praise and onset of animation on monitor).  Data suggested children with ASD spent 27% more time looking toward the robot administrator than the human administrator and did not fixate on either robot or target. Across all 48 trials with the robot, children with ASD successfully responded to the target hierarchy correctly in 95.83% of trials, a rate equal to TD success. In terms of intervention feasibility, we anticipated that many children would not tolerate wearing the LED cap; completion rates of 60% (ASD) and 80% (TD) were observed.  

Conclusions: Data indicated preschool children with ASD directed their gaze more frequently toward the humanoid-robot than human administrator and were capable of correctly responding to target.  As such, system enhancements successfully pushing toward correct orientation to target with systematically faded prompting and embedded coordinated action with human-partners could take advantage of baseline enhancements in attention and systematically incorporate more intrinsically interesting elements of the humanoid robot itself in order to enhance skills related to coordinated attention. Tolerance rates for wearing the LED hat highlight the need for the development of a non-invasive system for realistic extension on use with a young ASD population.