IEEE Robotics & Automation Magazine - June 2019 - 76

Table 2. HRI research on the transparency principle in technical terms (Continued ).
Study

Topic of Study

Transparency Out- Contribution
comes Investigated

Gutierrez, Chu,
Short, Niekum,
and Thomaz

Learning task de- Robot design to uncompositions for derstand humans
the construction
of generalizable
primitives and
task models

An intuitive interface
via keyframe demonstrations for teaching
robots to perform tasks

If the robot's task model conforms with
people's intuitions of task decomposition,
the robot's behavior will be more readily
understandable and aid in the update of
the robot's task models.

Hastie, Lohan,
Chantler, Robb,
Petrick, Lane,
Ramamoorthy,
and Vijayakumar

Increase transpar- Robot design to
ency to increase support human
trust in a robot's understanding
actions

Understanding the
user's mental model of
the system as a key to
providing just the right
information at the right
time

Balancing information presentation and
explanations with the user's actual needs
and cognitive load in a dynamic, fastmoving environment will be essential to
successfully deploy robotics and AI into
offshore robotics and elsewhere.

Racca and Kyrki

Challenges of
transparency for
learning robots

An example of the
enormous potential
of transparency to
improve the reliability
and performance of
learning robots

Exposure of the analytical and task models is a challenge (especially in black box
learning techniques). The modelling of the
human state is important, but research
is needed to develop methods to reliably
detect the human state and models to aid
the decision making of learning systems
during training.

Fischer

Transparency in
Negative outcomes
HRIs and its desir- and pitfalls of
ability
transparency, such
as misunderstandings and unnatural
interaction flow

Undesirable downsides
of transparency despite
its ability to enhance
explainability and predictability

Three downsides of transparency in HRI:
1) Transparency can destroy the illusion
that robots are similar to humans and
thus inhibit seamless HRI. 2) Transparency
about robots' high-level capabilities will
create assumptions that the robot also
has lower-level capabilities, which might
not necessarily be true and might create
misunderstandings. 3) Transparency about
low-level capabilities of a robot results in
unwanted inferences about the robot's
capabilities.

Broz, Ghosh,
Keller, McKenna,
Rajendran, and
Aylett

Design of a robot
for behavioral
skills training to
tackle autism
spectrum disorder in workplaces

Robot design to
support understanding of social
signals

System architecture
to enable behavioral
skills training through
a robot for individuals
with autism spectrum
disorder

A solid system architecture for behavioral
skills training for individuals with ASD
involves user-centered elements and
modelling of workplaces.

Kwon, Ferguson, Mann, and
Knepper

Implicit impressions of robot
competence

Transparency understood in terms
of warnings and investigation of their
effect on implicit
impressions

Measurement of
implicit impressions,
rather than explicit impressions, through the
affect misattribution
procedure

Transparency in the form of a warning did
not change competence judgment in the
short run but might have an effect in the
longer run.

Marmpena,
Dahl, and Lim

Human perception of robot
expressions and
emotions and
how they are
influenced by
transparency

Empathic and
anthropomorphic
weight attributed
to the robot by a
human

Linking transparency
and emotional perception: Does transparency
affect how robots are
perceived emotionally?

There were no empirical findings on the
transparency-emotions link, but a solid
taxonomy of emotional cues of robots and
their assessment in a two-dimensional
space (valence, arousal) was provided.

Wang, Barnes,
Pynadath, and
Hill

Automatic explanations of complex decisions

Trust in a robot, as
affected through
explanations and
team performance

Contrasting two types of
explanation: numeric information on uncertainty
and textual information
about sensor readings

Numerical information explanations were
seen as more open and competent, but
textual explanations were seen as more
predictable.

Oliveira, Arriaga, Display of steCorreia, and
reotype content,
Paiva
especially warmth
and competence,
by robots and
their effect

Robot design to
support human
understanding

Emotions, behavior, Applying the topic of
and intention for
stereotypes in a group
future interaction
interaction setting
with the robot

* See https://explainableroboticsystems.wordpress.com/.

76

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

JUNE 2019

Key Findings

No empirical findings were found at this
point.
https://explainableroboticsystems.wordpress.com/
IEEE Robotics & Automation Magazine - June 2019

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