IEEE Robotics & Automation Magazine - June 2019 - 86

design. For the making change experiment, a combination of
SLP and discrete and forward chaining was used. Unlike
the geometric assembly and AR experiments, the experimental design for the making change experiment used a

Diamond

Trapezoid

Hexagon
(a)

V-Shape

Two Houses
(b)

House

Tree
(c)

Butterfly

Ball

Figure 9. The puzzle tasks for the geometric assembly skills: (a)
subcomposition, (b) symmetry, and (c) complex assembly.

multiple-baseline-across-participants methodology to
control for any outside influences affecting the participants'
ability to make change.
An example prompt hierarchy for part of the making change
task is shown in Table 1. The SLP strategy was used, with
prompts arranged from least to most intrusive: verbal cue 1 is
the least intrusive; prompt level direction 2 serves as the controlling prompt. The student responses shown are as follows: no
response (NR), partially correct (PC), correct, and incorrect (I).
In addition, to evaluate the student volunteers' attitudes
about learning from a robot, Likert-type scale statements
and open-ended questions were used to collect subjective
data before and after participants interacted with the IRI
for the making change experiment. The results of this survey are discussed in the "Subjective Acceptability" section.
Intelligent AR Instruction Experiments
We also implemented response prompting for intelligent
instruction on a portable AR device. AR devices share similar
features with IRI instruction, such as context awareness, precision, and tirelessness when situated in an environment with
the user. Both are capable of delivering to students visual and
auditory prompts that interact with the real environment.
Using an AR device for instruction, we demonstrate that our
approach generalizes to systems with these overlapping features but a different physical implementation (i.e., a headmounted AR device versus a humanoid robot).
Figure 10 shows an overview of the AR system. When
wearing an AR device and learning a new sequential task, the
student can ask for help with the next step in the sequence at
any time. The AR device captures an image from the student's
point of view, processes it, and presents an appropriate
instructional prompt to the student via the AR device. In this

Table 1. The prompt hierarchy for the making change task using SLP.
Prompt Level

Response

Prompt Description

Verbal cue 1

Verbal interaction to determine how much change is due

Verbal encouragement, verbally providing goal

Differential positive reinforcement

Same as NR

Verbal interaction to determine which coin to begin with

Verbal encouragement, verbally providing goal and shortage between current state and
goal

Differential positive reinforcement

Verbal encouragement, providing goal and excess between current state and goal

Gesture to correct first coin, verbally providing goal

Gesture to correct next coin, verbally providing goal and shortage

Differential positive reinforcement

Gesture to coin to remove, verbally providing excess

Gesture to each coin to add, wait until added

Same as NR

Differential positive reinforcement

Gesture to each coin to remove, wait until removed, then same as NR

Verbal cue 2

Direction 1

Direction 2

IEEE ROBOTICS & AUTOMATION MAGAZINE

JUNE 2019

IEEE Robotics & Automation Magazine - June 2019

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