Professional Documents
Culture Documents
Perceived Appropriateness A Novel View For Remediating Robot Inappropriate Navigation Behavior in Narrow Environments
Perceived Appropriateness A Novel View For Remediating Robot Inappropriate Navigation Behavior in Narrow Environments
systematic literature review (including human-centered work as detected emotion and attention from social cues as additional input,
well as work on robotic agents) to survey the state of the art with and found that despite early exploration of mediocre performance,
respect to the following research question: What and how social detected emotion and attention still contribute to detection of per-
cues have been used and for what purposes in social navigation? ceived appropriateness (73.4%). Considering rich research already
For this purpose, we have collected and filtered 437 papers from in human emotion and attention detection, we expect that higher
sources of Scopus and Web of Science. Across two main application detection performance of perceived appropriateness can be easily
domains (navigation among pedestrians, and navigation among achieved.
vehicles), we identified three roles of social cues; informing pre-
dictions of future states, indicating internal states, and otherwise
signaling cue-related behavior.
Based on understandings of communications between agents, we
2.4 Effect of perceived appropriateness
found that implicit cues (trajectory, speed, head movements) ensure information on improving people’s trust
efficient communications, while explicit cues contribute to clarity of and acceptance of faulty robots
interactions. We see an increasing trend of robotic agents processing Our first exploration enable us to identify importance of emotion
and utilizing explicit cues, while perceived appropriateness convey and attention for detecting perceived appropriateness, yet is still
clearer information about people’s perception of the robot than far from good performance and ready-to-use. Yet we are interested
emotion and intention, which deserves more attention in the future. in how information of perceived appropriateness could be used
by robots to adapt its behavior and improve people’s trust and
2.2 Field observations of conflicts acceptance, thus higher social acceptance. Therefore, we plan to
To take a close look at perceived appropriateness, we first needed conduct a Wizard of Oz study.
a clearer understanding of conflicts, where perceived appropriate- The study will be carried out in narrow environments, in which
ness of robot navigation behavior typically arise. More specifically, the participant and the robot controlled by the operator will make
we needed to understand environmental and dynamical factors of about 20 encounters with each other. With participants’ self-reported
human-robot interactions that contribute to arising of conflicts, and perceived appropriateness, human operators will adapt robot behav-
how such conflicts are resolved by people to draw inspirations for ior for better human interaction experience. We will compare three
robot conflict resolving behavior. cases of interactions to identify importance of perceived appro-
We conducted a field study in which the robot controlled by priateness and its timing: with information, without information,
human operators navigate in social environments, and investigated and with delayed information. Participants will self-report their
how people respond and yield in conflicts with the robot. We found trust and acceptance [1] of robots when confronted with different
that narrow environments is an important factor for robot social cases of robot behavior, based on which results can be compared to
navigation, which triggers much stronger reactions and perception draw conclusions. We will also record human-robot interactions to
of the robot than spacious environments, yet has not been well analyze in more detail how human operators adapt robot behavior
studied. Besides, from our observations, we also found evidence to inspire robot behavior design.
suggesting strong correlations between emotion and perceived
appropriateness.
3 CONCLUSIONS
2.3 Data collection and analysis: Detecting
By reviewing of past literature in robot social navigation, we have
perceived appropriateness of robot social identified importance of perceived appropriateness for avoiding
navigation behavior repeated and continuous robot mistakes and improving robot so-
We thus took a closer look of human-robot interactions in narrow cial acceptance. We thus conducted a field study of conflicts and
environments, in which we designed 8 robot behavior to trigger observed rich perceived appropriateness especially in narrow envi-
a rich range of perceived appropriateness of robot social naviga- ronments and with specific robot behavior, where more attention
tion behavior. Based on understandings of appraisal theory [4, 5], is needed.
social signal processing approaches [3, 9], as well as our previous Based on these observations and past literature insights, we
observations, we made the assumption that emotion and attention conducted data collection of perceived appropriateness of robot
serve as important intermediate features for detecting perceived social navigation behavior, and proved emotion and attention as
appropriateness. With 31 participants, each interacting with all 8 important intermediates for detecting perceived appropriateness.
robot behaviors, we built a dataset of perceived appropriateness In the future, we plan to conduct a Wizard of Oz study to un-
of robot social navigation behavior (1008 datapoints). The dataset derstand how information of perceived appropriateness contribute
contains computed social cues and collected people’s self-reported to people’s trust and acceptance of erroneous robots, and draw
emotion, attention, and perceived appropriateness. insights for robot behavior design from flexibility and cognitive
Analysis of the dataset showed that emotion and attention con- decision making from human operators.
tribute most to detection accuracy of perceived appropriateness Our research contributes to understanding, detecting, and adapt-
(78.5%) compared with solely social cues (68.0%). Yet our emotion ing to conflicting interactions for robot social navigation, thus
and attention are collected based on people’s self-report, which is preventing robots from making repeated and continuous mistakes,
not accessible in real-life human-robot interactions. Therefore, we and increase their social acceptance.
Conference’17, July 2017, Washington, DC, USA