IEEE Robotics & Automation Magazine - June 2019 - 47

were the closest matches in age and gender to the ASD
participants. The mean age of this TD group was now
7.43 years (with a standard deviation of 2.29 years), with
all male participants.

similar trend is observed for the TD group despite multiple
outliers [Figure 9(b)]. Of course, for more accurate statistical
analysis, we will need to conduct a larger-scale study to verify
this finding.

Length of Interaction
In general, in the emotional-interaction activity, the TD
group was seen to take a larger number of turns than the
ASD group, as shown in Figure 8. The average length of
interaction for the TD group was 34.57 turns (with a standard deviation of 16.30 turns), while that for the ASD group
was 14.6 turns (with a standard deviation of 8.08 turns). The
one-way ANOVA for these data generated the following
results: F ^1, 11h = 6.57, p = 0.0263. Because p 1 0.05, we
concluded that there is a statistically significant difference
among the lengths of interaction for the two groups, with
the TD group taking part in longer interactions than the
ASD group.

Discussion
We developed an empathetic, emotionally expressive
robot-based tool to engage children with ASD in emotional interactions while simultaneously performing the
ER function. The novel,
consensus-based framework was tested with an
Results from the user study
emotionally interactive
agent-that is, the penconfirmed the feasibility
guin character-whose
emotional responses folof this framework as an
low the user's emotion
states closely (empathy)
ER tool that offers similar
before gradually guiding
them to the predeterbenefits to both groups of
mined goal state (ER).
Such empathetic behavchildren.
iors demonstrate the
capability of the framework to effectively model interaction patterns for diverse
personality types.
Results from the user study confirmed the feasibility of
this framework as an ER tool that offers similar benefits to
both groups of children. Although children with ASD
engaged in relatively shorter emotional interactions than did
TD children, we hope that continued interaction with these
tools in future sessions will lead to improvements in this
regard. The age ranges for the participants whose data were
analyzed in this study were five to 10 years for the TD group
and seven to 10 years for the ASD group. Therefore, while the
cognitive abilities of the two groups can be considered comparable, the within-group differences in emotional and language/reading abilities can perhaps be better addressed in a
future study.
For children in both groups, longer interactions in the
emotion game were seen to result in higher engagement
levels in the subsequent sensory station activity. This is
noteworthy because it points to the potential of this
framework as a tool to prime higher levels of focus and
engagement in various tasks for both groups. In testing
the application with participants, some obvious limitations were observed. We noticed that some of the younger
or lower-functioning children needed additional assistance in understanding the activity procedure, its purpose, and their role in it. In the future, we plan several
upgrades to this application, focusing particularly on
making the activity design and interface more intuitive
and engaging. We also intend to continue recruitment for
this study to close the age gap within and between the two
groups and to collect larger sets of data to improve the
reliability of our findings.

Success of ER
In terms of the ER function, our framework appeared to
perform with similar effectiveness for both the TD and
ASD groups. ER was successful for eight out of 11 participants in the TD group. However, out of the seven participants whose data were considered for analysis, successful
ER was achieved for four out of seven TD participants and
three out of six ASD participants. The one-way ANOVA
generated the following results: F ^1, 11h = 0.56, p = 0.471.
Because p 2 0.05, we concluded that there is no statistically significant difference between the effectiveness of the ER
function for the two groups. It can, therefore, be deduced
that the framework can effectively regulate the emotions of
both groups.
Effect on Engagement
Subsequent to engaging in the emotional-interaction activity with the penguin character, participants of this study took
part in another activity conducted with the two robots,
Romo and Mini, which was aimed at improving users' sensory processing skills. By comparing the two measures
described previously with the findings from this study, we
were able to identify a correlation among the performances
of the two groups in the emotional-interaction activity and
the sensory station activity.
Out of the 17 children who participated in the emotionalinteraction game, one ASD participant was unable to take
part in the sensory station activity. The mean age of this
group was now 8.2 years (with a standard deviation of 14.6
years). We plotted the lengths of interaction in the emotion
game with the penguin character against the engagement
index of the participants in the sensory station activity (Figure 9)
to identify any correlation.
For the ASD group, Figure 9(a) depicts a general increase
in the engagement index when the participant previously
experienced longer interactions with the penguin character in
the emotional-interaction game. Though not as clear, a

JUNE 2019

IEEE ROBOTICS & AUTOMATION MAGAZINE

IEEE Robotics & Automation Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2019