Understanding Interaction Dynamics in Socially Assistive Robotics with Children with ASD

Friday, May 15, 2015: 10:00 AM-1:30 PM
Imperial Ballroom (Grand America Hotel)
E. Short1 and M. J. Mataric2, (1)University of Southern California, Los Angeles, CA, (2)Computer Science, University of Southern California, Los Angeles, CA
We present a novel analysis of an existing dataset examining one-on-one interactions between children with an ASD diagnosis and a robot. We conclude that children's reaction to the robot can vary on a continuum of object-to-agent, at least partially independent of the designers’ intent. We also find that some behaviors meant to promote social responses may inadvertently inhibit them by promoting other, conflicting behaviors.


An increasing body of work in human-robot interaction for socially assistive robotics (SAR) for children with ASDs points towards the value of SAR as an intervention tool (Scassellati, Admoni, and Matarić, 2012).  Our work contributes to this area by extending our understanding of children’s behavior using dynamics of the interaction to understand children’s perceptions of robot agency.


Based on a qualitative evaluation of the video data from the study, we defined the following broad hypotheses:

  1. For the children who spoke most, the robot’s bubble-blowing distracted from the interaction;

  2. For the children who spoke least, the robot’s bubble-blowing engaged them in the interaction;

  3. Differences in behavior were more a function of individual reactions than the implemented differences between the robots and can be classified into agent-like vs. object-like treatment of the robot.


The data discussed in this abstract come from a prior experiment described in Feil-Seifer and Matarić (2012); the analysis and results described are new.  The data were hand-coded for 20 robot, child, and parent behaviors.  In order to model the dynamics of behavior, we examine ordered pairs of events (first A happened, then B).  In addition, we use a clustering algorithm to group sessions based on participants’ behavior.


We find limited differences in the behavior of children across the original experimental conditions.  However, we find correlations between their overall speech and their reactions to the robot's bubble-blowing behavior, suggesting that the behavior is not beneficial with more verbal children, while it can be a way to engage less verbal children.  Our clustering uncovers a more meaningful classification based on the child-robot interaction pattern: while some children are highly engaged with the robot verbally (as if with an agent), other children are more engaged with the button pushing or movement behaviors (as if with an object).

Conclusions:  One of the most exciting aspects of SAR for children with ASDs is the potential for robots that can support therapeutic goals. We suggest the division of robot behaviors into agent-like and object-like, and the customization of the robot’s role to an individual child's needs.


Feil-Seifer, D., & Matarić, M. (2012). Distance-Based Computational Models for Facilitating Robot Interaction with Children. Journal of Human-Robot Interaction.

Guthrie, D., Allison, B., Liu, W., Guthrie, L., & Wilks, Y. (2006). A Closer Look at Skip-gram Modelling. In Proceedings of the Fifth international Conference on Language Resources and Evaluation. Genoa, Italy.

Scassellati, B., Henny Admoni, & Matarić, M. (2012). Robots for use in autism research. Annual Review of Biomedical Engineering, 14, 275–94.