IEEE Robotics & Automation Magazine - June 2019 - 75

Table 2. HRI research on the transparency principle in technical terms.*
Study

Topic of Study

Transparency Out- Contribution
comes Investigated

Key Findings

Holder and
Marge

Role of intent in
HRI to improve
explainability

Robot design to
support human
understanding

Hellström and
Bensch

Meaning of
understandability
and its formalization

Human understand- Model of interaction for There is a conceptual separation of
ing of robot's beunderstanding
information to be communicated from the
havior; robot design
means of communication.
to support human
understanding

Taxonomy for intent in An intent statement needs additional
HRI in a military context background information and is contextdependent.

Gong and Zhang Reciprocal under- Human understandstanding between ing of robot's berobot and human havior; robot design
to support human
understanding

Formulation of human
interpreting robot's
actions as a labeling
process

Thellman and
Ziemke

Prediction and
explanation of
robot actions

Observation, ascription, People's predictions and explanations are
and inference model
constrained by theoretical presuppositions
regarding the robots' role-specific goals,
morphological and environmental constraints, sensory capabilities, and situated
perspective on the world.

Avrunin,
Rosenthal, and
Simmons

Appropriate level Robot design to
of detail in expla- support human
nations
understanding

Human understanding of robot's
behavior

Follow the label process: check if the
action is explainable; if not, search for
optimal timing and content to signal its
intentions.

Coverage-based
explanation-reduction
algorithm

Robots can create reduced explanations
that approximate the true robot policy but
are more memorable or understandable
to a human.

Bekele, Lawson, Human-level
Robot design to
Horne, and
explanatory
support human
Khemlani
biases for explain- understanding
ability (in-person
reidentification)

Creation of a multiattribute residual network
for explanatory reidentification

Deep learning systems capable of mimicking human explanatory biases can provide
meaningful and interpretable explanations
of their own internal operations.

Huang, Bhatia,
Abbeel, and
Dragan

Establishing trust Robot design to
via critical states support human
understanding

Computation and use
of critical states

The end user does not need to know what
the robot would do in all states; instead,
the robot action matters only in critical
states. Showing end users how the robot
acts in critical states gives them a better
understanding of what it has learned and
enables them to decide in which situations they can trust the robot.

Korpan and
Epstein

Natural explanations of robot
navigation plans

Robot design to
support human
understanding

Method that compares
the perspectives of an
autonomous robot and
a person when they
plan a path for navigation

By explaining the context of the robot's
most recent action and its long-range
perspective, robots can produce meaningful, human-friendly explanations quickly in
natural language.

Chiyah Garcia,
Robb, Liu, Laskov, Patron, and
Hastie

Natural language Robot design to
interface for
support human
remote autonomy understanding
explanations

Model to allow ondemand queries for
status and explanations
of behavior

If the expert is from the same pool of end
users (i.e., operators), explanations are
likely to align with their mental models
and assumptions about the system.

Ghayoumi

Cognitive-based
emotion model
for social robots

Architecture for emotion and social robot
model incorporating a
knowledge-based system and the cognitive
appraisal

Applying this model in health care allows the robot to report emergency and
nonemergency cases to the experts and
communicate with patients with a proper
expressive face.

Erel, Hoffman,
and Zuckerman

Interpreting
Human undernonanthropomor- standing of robot
phic robots' social behavior
gestures

Gestures designed by
performance artists
and movement experts
and purely mechanistic
gestures, suggesting
specific loci of focus for
gesture design

Limited gestures of nonanthropomorphic
robots can be consistently interpreted as
social interaction cues. Specific movement components (e.g., vertical axis) have
more profound effects on the emotion
perceived from the robot.

Robot design to
support human
understanding

(Continued )

JUNE 2019

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

75
IEEE Robotics & Automation Magazine - June 2019

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