Comparing Autistic Children’s Social Communication Behaviours in a Robot-Assisted Versus Adult-Led Activity

Background: Autistic children experience difficulties with social communication behaviours (SCB). Robots have been proposed as potentially useful tools to support autistic children’s SCB, based on claims that their predictable behaviour and simplified characteristics might be less socially demanding for autistic children than human interactions. Yet, very few studies have directly compared autistic children’s SCB when interacting with robots versus humans.

Objectives: The DE-ENIGMA project is a large, collaborative project investigating the potential of humanoid robots as tools in autism education, including for teaching about social behaviours. Here, we directly compared autistic children’s use of SCBs in child-robot versus child-adult interactions.

Methods: An initial feasibility study collected recordings from a large cohort (n=128) of 5- to 12-year-old autistic children in special education settings, in the context of a 6-step emotion-recognition teaching programme (based on Howlin et al., 1999). Children were randomly assigned to robot-assisted or adult-led teaching conditions. We selected a subset of 5- to 8-year-old participants with at least 15 minutes of recorded interaction during the feasibility study for annotation and analysis. Groups were well-matched for age (p=.699) and for autism severity symptoms, as measured by the CARS2-ST (p=.485). We designed a coding scheme based on the SCERTS model (Prizant et al., 2006) to identify a range of children’s SCB in the video data, including commenting; pointing; imitation; protesting; requesting; responding, etc. Instances of shared affect were also annotated.

Results: Data annotation and analysis are ongoing. Of the data that have been annotated (n=12; 2 females; 6 children from each condition), there were 926 instances of SCBs (excluding shared affect) in a 180-minute sample (15 minutes/child). The median number of SCBs in the robot-assisted condition was 59.50 (range = 17 – 87), while in the adult-led condition was 84.50 (range= 35 – 198), although there was no significant difference between teaching conditions (Mann-Whitney U=24.00, p=.394, r=.28). When comparing the medians of each type of SCB, we found a significant difference for imitative behaviours between the adult-led condition (Mdn=1) and the robot-assisted condition (Mdn=14.50), U=33.50, p=.012, r=.72. There were no significant between-group differences for any other SCBs.

Shared affect was frequently combined with other behaviours; it is reported separately to give a clearer picture of its extent (Figure 1). We found a significant difference for sharing positive affect between the robot-assisted condition (Mdn = 10.5) and the adult-led condition (Mdn = 3), U=5.50, p=.045, r=-.58. Sharing negative affect did not significantly differ between teaching conditions (U=9.00, p=.059, r=-.55).

Conclusions: Of the data examined thus far, the results indicate that children’s SCBs were similar in the robot-assisted and adult-led teaching conditions on the total number of SCBs observed, and on instances of most individual SCB types. It is encouraging that children shared more positive affect in the robot-assisted condition, indicating that a humanoid robot can be part of enjoyable, social interactions. Variance in children’s SCBs within each group highlights the importance of considering individual differences when studying social communication in autistic children, particularly those with limited spoken communication.