More about HKUST
Understanding and Supporting Human Feedback Elicitation for Human-Robot Interaction
The Hong Kong University of Science and Technology
Department of Computer Science and Engineering
PhD Thesis Defence
Title: "Understanding and Supporting Human Feedback Elicitation for
Human-Robot Interaction"
By
Miss Hanfang LYU
Abstract:
Robots are increasingly deployed in everyday human-centered spaces, where
aligning their behavior with human values and situational needs relies on
high-quality human feedback, i.e., human signals that inform or correct robot
behavior, including preferences, social cues, and corrective actions.
Eliciting such feedback efficiently, however, remains non-trivial in both
directions of human-robot communication. On the understanding human side,
existing work tends to treat human feedback as homogeneous, well-formed
input, and does not yet offer a fine-grained account of how its form and
modality shift across the heterogeneous interaction contexts in which it
arises. On the robot design side, current systems and robot behaviors do not
sufficiently support users' actual feedback behavior, since non-expert users
often do not know how to translate their preferences or natural expressions
into robot-readable signals, while their motivation is further bounded by
tedious labeling tasks, occupied cognitive resources during competing
activities, and the limited sympathy that machine-like robots evoke when they
encounter errors.
This thesis addresses both gaps across three contexts of post-hoc trajectory
labeling, in-situ social multitasking, and reactive error moments, where each
work first deepens the understanding of how humans express feedback in that
context, and then translates the findings into supportive system or behavior
design. For the post-hoc context, a formative study (n = 12) on
preference-based reward learning characterizes three labeler challenges in
criteria formation, detail noticing, and focus maintenance. The resulting
FARPLS system, equipped with feature-clustered prompting, keyframe markers,
and real-time attention monitoring, significantly improves labeling
consistency and engagement without raising cognitive load (n = 42). For the
in-situ context, an elicitation study (n = 24) in an AR-simulated coffee-chat
scenario investigates how participants signal 13 intentions and feedback acts
to four service robots of distinct morphologies (anthropomorphic, zoomorphic,
grounded technical, and aerial technical) while acting as a speaker or a
listener. Coding of 624 social cues with retrospective think-aloud interviews
reveals that conversational role shapes verbal and gestural cue choice
whereas morphology shapes modality choice via perceived sensing, yielding
design implications for low-disruption, role- and morphology-aware
service-robot response. For the reactive context, co-design workshops (n =
21) and member-checking interviews (n = 18) produce human- and
animal-inspired expressive error reactions for machine-like robots, surfacing
a gap between designer intent and observer perception together with the
conditions that evoke sympathy and helping intention. We implemented human-,
animal-, and machine-like designs on an interactive WoZ-controlled MR
platform that embeds them within a continuous robot task workflow, enabling
real-time operator control while participants interact with shared household
objects.
In summary, this thesis contributes to human-robot interaction by delineating
an understanding of how humans provide feedback across the post-hoc, in-situ,
and reactive interaction contexts, by developing novel interfaces, behavior
design strategies and system prototypes that support feedback elicitation,
and by offering empirical insights from formative, elicitation, and co-design
studies, that inform the robot designs to support human feedback elicitation.
Date: Friday, 31 July 2026
Time: 10:00am - 12:00noon
Venue: Room 3494
Lifts 25-26
Chairman: Dr. Jiying LI (OCES)
Committee Members: Dr. Xiaojuan MA (Supervisor)
Prof. Fugee TSUNG (Co-supervisor, IEDA)
Prof. Andrew HORNER
Dr. Arpit NARECHANIA
Dr. Yantao YU (CIVL)
Dr. Kening ZHU (CityU)