IEEE Robotics & Automation Magazine - June 2019 - 22

the course of the different reward phases could be due to
external factors such as learning or fatigue. This result supports the idea that infants may be trying to imitate the NAO
robot, although more research is necessary to further test the
imitation hypothesis.
Parental perceptions support some of the additional
understanding described in H3. Although parental perceptions of which reward motivated their children to
move most did not perfectly match the top rewards suggested by the contingency learning results, parents did
perceive that their infants were moving slightly more than
normal. Overall, parental perceptions of the interaction
were very positive. This is important for the promise of
SAR adoption in infant interventions. Additionally, parents were happy with the challenge level of the activity,
which may indicate that the difficulty of the study design
was appropriate for the participant population. Another
encouraging note was that parents positively perceived
socially assistive robots, although none of them had prior
experience with robots.

Infant Engagement

Rating

8
7

∗

6
5
+
+

4
3
2
1
0

+
Movement

Movement
and Lights

Movement
and Sound

Rating

Study Phase
(a)
8
7
6
5
4
3
2
1
0

Infant Fun
∗

Movement

Movement
and Lights

Movement
and Sound

Study Phase
(b)
Figure 7. The parents' ratings of the infants' engagement and
fun levels during each reward condition. The center box lines
represent the median, and the box edges are the 25th and 75th
percentiles. The whiskers show the range up to 1.5 times the
interquartile range, and the outliers are marked with a "+." The
brackets indicate significant differences as determined by the
post hoc multiple comparisons test.

IEEE ROBOTICS & AUTOMATION MAGAZINE

JUNE 2019

Major Strengths and Limitations
A major contribution of this research is that it integrates
SAR into a new and potentially high-impact domain: teaching and reinforcing infant motion. Although studies involving infants can be challenging because of a variety of
factors, we were able to recruit the targeted number of participants for this research. The infants were engaged by the
robot and remained alert for the vast majority of the interaction time. The results of this first study in socially assistive infant-robot interaction suggest that infants can learn
robot-delivered contingent rewards and may try to imitate
a robot.
The designed SAR system can be used to deliver a variety
of rewards for a nuanced range of infant motions and interventions. As discussed in the "Related Works" section, the
robot's physical embodiment and ability to provide a variety
of reward types help it to motivate infants and hold their
attention for longer than other therapeutic tools. The infantlike size and humanoid anatomy of the robot also allow it to
fit in the infant's visual field and demonstrate motions that an
infant can imitate. This observation, combined with other
study results, provides us with novel insights that will inform
future targeted SAR interventions for infants.
Because this data collection involved very young infants,
we limited the session duration to minimize the possibility of
infant stress or fatigue. We were also limited in the number of
conditions we could include in the experimental design; additional types of demonstrations and rewards may help us learn
more about how to teach and reinforce infant motion. The
US$9,000 cost of the NAO robot is high, but this price is competitive with medical equipment, and the robot is more dexterous and versatile than past contingency learning tools, e.g.,
mobiles or toys. Our software architecture is designed to
allow for replacement of the NAO robot with a more costeffective alternative when one becomes available. Finally, our
study involved a single-session design in a laboratory setting.
In the future, data gathered in a natural environment over a
longer interaction period will further inform our work on
SAR for infants and increase the potential benefit of SARbased infant interventions.
Key Contributions and Future Work
The results of this article further the general understanding of
infant-robot interactions and build the foundation for a new
research area focused on using SAR to teach and reinforce
infant motion. In this article, we focused on six- to eightmonth-old infants' responses to a small humanoid robot. Specifically, we investigated how frequently infants produced leg
movement above an acceleration threshold when being
rewarded by robot behaviors, whether infants imitated the
robot's ball-kick motion, and how parents perceived the
infant-robot interactions. We found that the majority of
infants learned the contingency, most infants imitated the
robot in at least one of the reward phases, and parental opinions of the robot were generally high, although their perceptions of infant motion levels and robot imitation were only

IEEE Robotics & Automation Magazine - June 2019

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