Kinematic Performance during a Human-Robot Imitative Gesturing Task Differentiates Autism Spectrum Disorder (ASD) from Typical Development and Correlates with Clinical Assessments of Motor Skills

Background: Individuals with Autism Spectrum Disorder (ASD) may have difficulty imitating social-communicative gestures, perhaps in part due to significant challenges in areas of motor control including postural stability and coordination. Currently, assessment of a person’s ability to appropriately use gestures relies on subjective observation and parent-report. While useful clinically, these approaches do not identify kinematic signatures of ASD and typical development (TD) that may enhance our understanding of the underlying mechanisms driving imitative gesturing differences. Technologies such as motion-capture and robotics may help to quantify motor problems in ASD.

Objectives: We aimed to quantify differences in imitative gesturing between ASD and TD using motion capture during human-robot interaction. We also aimed to determine whether scores on common clinical measures of motor function and symptom severity capture the full scope of imitation difficulties in ASD.

Methods: Thirty-two participants with ASD (n = 18) and TD (n = 14) imitated unilateral and bilateral social gestures of an interactive robot (e.g., wave). Participants and the robot were instrumented with reflective markers on corresponding head and body locations. Body position during imitation was tracked by an infrared motion-capture system. We used Dynamic Time Warping (DTW) to quantify the degree to which the participant’s and robot’s movement were aligned across the movement cycle. Participants also completed a battery of social, motor, cognitive, and behavioral assessments.

Results: We used repeated-measures ANOVA to test main and interaction effects of gesture type (fist bump, give, hug, wave, ask, celebrate) and group (ASD, TD) on imitation accuracy (DTW cost), including age as a covariate in the model. The main effect of gesture type was significant (F(5, 145) = 37.92, p < .001) with all gestures differing significantly from one another (ps < 0.005) except for “celebrate” and “wave”. The main effect of the covariate, age, was significant (F(1, 29) = 5.27, p = 0.29,η_p² = 0.15), but its interaction with gesture was not (p > 0.05). The main effect of group was nonsignificant (F(1, 29) = 2.11, p = 0.16), with low observed power (0.29). The interaction of gesture and group was significant (F(5, 145)= 2.56, p = 0.03, η_p² = 0.08), with the ASD group demonstrating significantly lower accuracy for the “fist bump” and “celebrate” gesture types than the TD group. Imitation accuracy was correlated with MABC-2 aiming & catching, balance, and total scores for both the “fist bump” and “celebrate” gestures (ps < 0.05), but not ADOS-2 scores.

Conclusions: Preliminary results revealed kinematic differences between ASD and TD in imitative accuracy, which related to clinical motor assessments. Future analyses will include examination of positional matching for specific joint angles (e.g., elbow flexion/extension), as well as correlational analyses for additional clinical tests. Imitative gesturing is a building block to later development of social-communication skills; difficulty with reproduction and functional use of gestures may negatively impact social engagement, and in turn, learning opportunities. It is important to understand the underlying motor control and sensorimotor integration mechanisms that support imitative gesturing in ASD in order to identify appropriate intervention targets.