IEEE Robotics & Automation Magazine - June 2019 - 15

Table 1. The study participants' demographic information.
Infant

Gender

Age
(Days)

AIMS
Score

AIMS
Percentile

Weight
(kg)

Head Circumference
(cm)

Body Length
(cm)

TD2

183

6.6

TD3

258

7.28

45.5

TD4

264

TD5

211

10.1

TD6

226

7.05

43.5

TD7

195

6.66

43.5

67.5

TD8

191

7.57

TD9

194

8.5

43.5

TD10

183

6.23

41.5

TD11

182

8.08

TD12

206

7.8

TD13

188

7.2

AIMS: Alberta Infant Motor Scale.

parental perceptions of the child-robot interaction compare
to measures of infant motion and behavior. The experimental
procedures described throughout this section were approved
by the University of Southern California Institutional Review
Board under protocol #HS-14-00911.
Participants
We recruited 12 typically developing six- to eight-month-old
infants from the greater Los Angeles area to participate in
our study. We selected the six- to eight-month age range
because infants can learn contingencies starting before six
months of age (e.g., [16]) and six months of age is a common
infant age for the initiation of related work assessing the type
of movement behavior we are studying (e.g., [6]). Table 1
displays the age, size, and developmental information for
each infant.
Study Setup
Based on the literature discussed in the "Related Work" section as well as our pilot study results [11], we chose the
Aldebaran NAO humanoid robot for our infant-robot interventions. In the experimental setup, the NAO robot and
infant sat facing one another in a small room with white walls
and minimal visual distractions, as shown in Figure 2. The
chair where the infant sat allowed for full leg mobility. Infants
wore APDM Opal inertial sensors on both arms and legs so
that we could measure the triaxial acceleration and angular
velocity of each limb.
Infant participants also wore a head-mounted eye tracker,
and three red-green-blue (RGB) cameras and a Microsoft
Kinect One RGB-depth (RGB-D) camera captured front, side,
and face views of the infant. The setup included two suspended
toy balls, one that the robot could kick with its left leg and the
other that the infant could kick with his/her right leg. This

Eye Tracker

Inertial Sensors
Figure 2. The experimental setup. The infant interacts with an
NAO robot while the labeled sensors (an eye tracker and inertial
sensors) and additional sensors (RGB cameras and a Kinect One
RGB-D sensor, which are not shown in the field of view of this
image) capture information about the infant-robot interaction.

object setup was informed by past work showing that instrumental behavior (e.g., kicking a ball) motivates infants more
than spontaneous behavior (e.g., kicking for the sake of
kicking) [18].
Manipulated Variable
In addition to learning how to encourage infant motion, a key
goal of this study was to determine what types of robot
rewards would be most effective for encouraging infant
motion. Accordingly, the manipulated variable in this study
was the type of contingent reward given. In the within-subjects study design, each infant experienced three types of contingent rewards in three separate phases. To avoid ordering
effects, the condition order was counterbalanced and randomly assigned to participants. The three reward types for
achieving leg movements above a previously determined
acceleration threshold [20] were the following:
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