AI models may make robotic interactions much more beneficial for children with an autism spectrum disorder (ASD), according to new research published in Social Robotics.
Approximately one in 59 children in the United States are diagnosed with an ASD, the authors noted, and socially assistive robotics (SAR) is a type of human-robot interaction (HRI) that could potentially help those patients develop social skills—but more research is still needed in this area.
“HRI methods are still limited in their ability to autonomously perceive, interpret, and naturally respond to behavioral cues from atypical users in everyday contexts,” wrote lead author Shomik Jain, University of Southern California in Los Angeles, and colleagues. “This hinders the ability of SAR interventions to be tailored toward the specific needs of each user.”
Jain et al. explored the possibility of using supervised machine learning techniques to optimize SAR interventions for children with an ASD. Data was collected from one month of in-home SAR interventions with seven child participants. The team surveyed when the children were engaged by the interaction or disengaged “using standard definitions of engagement as a combination of behavioral, affective and cognitive constructs.”
Overall, the team developed two supervised machine learning models—one that ran on data from numerous users and one that created an individual model for each user. The average area under the ROC curve was roughly 90%, and the team said its models were “feasible for use in online robot perception of and closed-loop response to disengagement.” The recall for disengagement, however, was just 50%, “a key area of future work.”
“The presented models are also readily interpretable, an important characteristic for facilitating implementation,” the authors wrote. “Interpretability of machine learning is especially important in the ASD context, where therapists and caregivers need an understanding of the system’s behavior to trust and adopt it.”
Jain and colleagues did note that there was a significant amount of variance in their data from one child to the next, both because parents/caregivers would adjust the camera at various times and because recorded audio “contained a high level of background noise.” This variance made the AI models “vulnerable to overfitting,” and “bagging, boosting and early stopping” were used to combat that problem.
“Overall, online recognition of and response to disengagement will enable the design of more engaging, personalized, and effective HRI, especially in SAR for the ASD community,” the authors concluded.
Michael has more than 16 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.