IEEE Robotics & Automation Magazine - June 2019 - 56

autism (85%). During the study, among the 416 subjects, 22%
were parents of children with ASDs and 16% were therapists
or teachers of children with ASDs. They were surveyed with
questions such as "Is it ethically acceptable that social robots
are used in therapy for children with autism?" or "Is it ethically acceptable to use social robots that replace therapists for
teaching skills to children with autism?" This survey indicated
the importance of stakeholder involvement in the process
with a focus on specific health-care issues.
In another study developed within DREAM, Peca [10]
explored whether age, gender, education, previous experience
with robots, or involvement with persons with
ASDs influences people's
In terms of social
attitudes about the use of
robots in RET. Results
engagement, the children
show that these socialdemographic factors have
showed more interest
a relevant impact on how
social robots are perin the robot partner
ceived, e.g., men seem to
have a higher level of ethifor the duration of the
cal acceptability compare
to women, younger parintervention.
ticipants seem to be more
open to accepting the use
of social robots in RET for ASDs compared to older participants. In terms of the involvement of the participant with
children with ASDs and the use of social robots in RET, the
study suggests that parents who are not involved directly with
ASD children have a higher ethical acceptability level than
those who are directly involved.
Finally, Richardson et al. [26] have addressed a debate that
discusses the risks and challenges of developing research by a
multidisciplinary research team with a vulnerable population,
such as children with autism. Given the different backgrounds,
research goals, assumptions, and practices, each multidisciplinary research team would approach the research topic from
different perspectives, i.e., experimental, clinical, engineering,
philosophical, and anthropological. Each discipline has its own
history, terminology, methods, and preferences, therefore, synthesizing these approaches can be challenging.

Figure 9. The control panel on a teacher's tablet.

IEEE ROBOTICS & AUTOMATION MAGAZINE

JUNE 2019

Discussion and Conclusions
With the DREAM project, we attempted to implement RET
in children with autism interventions. In this article, we have
highlighted the technical development and clinical validation
of this approach.
Given the sensitive environment where RET is utilized, the
DREAM system was developed by taking into account the
requirements from both therapeutic and robotic perspectives
(see the "Requirements for RET Systems" section). The supervised autonomous system follows a multilayered behavioral
organization for generating task-based and social behaviors. It
was engineered by following a modular approach so that, along
with being an open source software, the system may be easily
used, adapted, and/or extended by other research teams for use
with different therapeutic scenarios and robotic platforms.
Our system reaches a performance on par with human
therapies commonly used today in clinical studies. Despite
the mixed results obtained during the single-case experiments, these studies offered valuable insights into the variability of the response of ASD children to RET and pointed
to some important issues that should be accounted for when
developing such interventions (e.g., the need for personalized interventions that match each child's skill level). Further exploration must be done using variables involved in
outcomes, such as social engagement, positive and negative
emotions, adaptive and maladaptive behaviors, and rational
and irrational beliefs [27]. These variables have previously
been studied for the first phase of the clinical trial. In terms
of social engagement, the children showed more interest in
the robot partner for the duration of the intervention. Positive emotions appeared more often while interacting with
the robot during the imitation and joint-attention tasks. The
presence of the robot usually acts as a behavioral activator,
so that both adaptive and maladaptive behaviors seem to
appear more often in the WoZ condition compared to the
SHT condition. The same study is currently being done for
the second phase, and it will offer better answers regarding
the relative efficacy of RET and ASD interventions. Future
investigations should aim for new research questions other
than determining whether RET is more or less effective than
standard treatments, e.g., is RET faster than therapist-mediated interventions and which children could benefit the
most from RET interventions and under which conditions.
As a final clinical conclusion and supported by data
obtained during the experiments, we can say that RET is a
promising approach that could be as efficient as (or even
more efficient than) classical interventions for a large variety
of outcomes for children with ASDs.
Given its technical requirements, unfortunately, very few
end users may benefit from the system developed in this project. For that reason, a simplified version of the DREAM system has been implemented as one of the Ask NAO [31]
Tablet applications (Figure 9). Its functionality is as follows.
The caregiver uses the application as an administrator (which
allows him/her to monitor NAO's activities and access the
control panel), and the child uses another tablet that can

IEEE Robotics & Automation Magazine - June 2019

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