Humanoid Robot Social Intervention Shows Promise

Background: Affect recognition is a core deficit in autism spectrum disorder (ASD); this deficit compromises ability to interpret facial cues to interpreting others’ communicative intent. Humanoid robots offer potentially useful supports for interventions targeting social skills of children with ASD.

Objectives: 1) To evaluate responsiveness of children with ASD to a robot-mediated affect recognition intervention (ARI); and 2) To gather preliminary evidence on effect of an ARI curriculum on distal measures of affect recognition.

Methods: An RCT was used to investigate the effects of a social skills ARI for children 4-8 years with ASD on distal affect recognition measures. Participants were randomized to the ARI (Milo) (n=20) or Treatment-As-Usual (TAU) group (n=20) utilizing a matched-pairs design. The Milo group participated in a maximum of 24 sessions with the Milo robot and an instructor. Three treatment subgroups were created based on participants’ rate of skill acquisition in Milo’s affect curriculum (Slower, Moderate, Average+Rapid Rate). Demographic and outcome variables were examined for each treatment subgroup.

Results: Milo participants advanced through the ARI curriculum with an average total accuracy rate of 84%. Families were highly engaged, reflected by low attrition (5%) during the treatment phase. All parents either agreed (16%) or strongly agreed (84%) on the importance of weekly Milo treatment sessions, and 79% of parents rated the treatment as “very” or “very much” worthwhile. In the Milo group, 70% of children reported enjoying learning about emotions, and 75% reported enjoying how the robot helped them learn emotions. Descriptive analyses of the Milo subgroups revealed that Slower Rate (M age=55.71 months, SD=5.85) and Moderate Rate (M=59.46 months, SD=18.70) groups were significantly younger than the Average+Rapid Rate (M=81.71 months, SD=9.35) group (p=.0004, p=.0065). The Slower Rate group had higher ADOS-2 Total scores (M_Total=19.33, SD=1.15) and calibrated severity scores (CSS) (M_CSS=9.00, SD=1.00) compared to the Moderate (M_Total=9.66, SD=3.05; M_CSS=5.00, SD=1.00; p=.0068, p=.0080, respectively) and Average+Rapid Rate groups (M_Total=12.92, SD=4.64; M_CSS=7.35, SD=1.69; p=.0351,p=.1310, respectively). Repeated measures ANOVAs revealed no significant differences between Milo and TAU groups on the NEPSY-II Affect Recognition subscale [F(1,37)=0.05, p=0.818] or on The Transporters computerized quiz (Autism Research Centre, 2009) (1) matching questions [F(1,37)=0.18, p=0.6749], (2) identification questions [F(1,37)=1.63,p=0.2097], and (3) scenario questions [F(1,37)=0.16,p=0.6928].

Conclusions: Results show promise for humanoid robot-mediated intervention in engaging children with ASD. Indicators of treatment acceptability were high among parents and children, suggesting social validity of the intervention. A comprehensive analysis of the three Milo subgroups revealed important and unique trends that may inform future research with children with ASD and ARI. Older children and those with lower levels of ASD symptoms tended to progress through the curriculum at a faster rate than younger children or those with more severe ASD symptoms. Exploratory findings suggest there may be specific variables that could predict children’s responsiveness to social robotic interventions. Although group differences were minimal on distal measures of affect recognition, the Milo robot has several other social skills curricula which warrant evaluation of efficacy.