Identifying Attentional Differences in Children with and without ASD: A Human-Robot Interaction Study

Background: The ability to share one’s focus of attention with another person is essential for social learning during early development (Carpenter et al., 1998). Children with autism spectrum disorder (ASD) show impairments in joint attention, specifically its initiation (Mundy & Newell, 2007). Typically, the presence or absence of such impairments in children is ascertained by trained clinicians in carefully planned play sessions during the diagnostic decision-making process (Dawson et al., 1998). Although robots are more commonly viewed as an interventional tool for children with ASD (Scassellati et al., 2012), they also have the potential to provide clinically relevant markers to aid in behavioral phenotyping (Scassellati, 2005).

Objectives: To examine behaviors related to the initiation of joint attention (IJA) in toddlers with ASD and with no clinical diagnosis (typical development; TD) during a short, unstructured interaction with a robot.

Methods: This study involved 22 toddlers (n_ASD = 11; n_TD = 11) matched on chronological age. Participants ranged in age from 20 to 42 months (M= 30.8 months). The human-robot interaction lasted approximately ten minutes and took place in an assessment room with a research assistant and typically a caregiver present. The spherical robot used light, sound, and movement to emulate the emotions of joy, fear, anger, and sadness in a fixed pattern that was at times contingent upon child interaction (Boccanfuso et al., 2015). No restrictions on physical contact were placed. The research assistant engaged the child by asking questions about the robot, but refrained from attributing emotions to it. Percentage of the total interaction time was calculated to assess duration of gaze behaviors and physical contact, while frequency counts were used to measure gesturing behaviors, instances of cooperative play, and gaze alternation between the robot and an adult.

Results: A series of one-way ANOVAs revealed differences between IJA behaviors in the two groups during the robot interaction. The group of children with ASD spent significantly less time looking at others’ faces during the interaction than the TD group (F(1,20) = 13.13, p < .01, d = -1.55). Similarly, attempts to share attention by alternating gaze from the robot to an adult and vice versa were significantly lower for toddlers with ASD than TD toddlers (F(1, 20) = 6.61, p < .05, d = -1.10). There was a marginally significant difference between the cooperative play behaviors of the two groups (ASD < TD; F(1, 20) = 6.61, p = .058, d = -.86). No significant between group differences were found for the percent of total interaction time spent looking at the robot, time spent in physical contact with an adult, frequency of pointing behaviors at the robot, or frequency of gestures toward other items in the assessment room.

Conclusions: Our results highlight the differences between children with and without ASD during a brief, unstructured play session with a robot and adult supervisors. The findings are consistent with previous research on IJA, and they suggest a potential auxiliary role for robotics in the early identification of ASD.