Available via license: CC BY 4.0
Content may be subject to copyright.
Effect of Adapting to Human Preferences on Trust in Human-Robot Teaming
Shreyas Bhat1, Joseph B. Lyons2, Cong Shi3, X. Jessie Yang1
1University of Michigan
2Air Force Research Laboratory
3Miami Herbert Business School
shreyasb@umich.edu, joseph.lyons.6@us.af.mil, congshi@bus.miami.edu, xijyang@umich.edu
Abstract
We present the effect of adapting to human preferences on
trust in a human-robot teaming task. The team performs a
task in which the robot acts as an action recommender to
the human. It is assumed that the behavior of the human
and the robot is based on some reward function they try to
optimize. We use a new human trust-behavior model that
enables the robot to learn and adapt to the human’s pref-
erences in real-time during their interaction using Bayesian
Inverse Reinforcement Learning. We present three strategies
for the robot to interact with a human: a non-learner strategy,
in which the robot assumes that the human’s reward func-
tion is the same as the robot’s, a non-adaptive learner strat-
egy that learns the human’s reward function for performance
estimation, but still optimizes its own reward function, and
an adaptive-learner strategy that learns the human’s reward
function for performance estimation and also optimizes this
learned reward function. Results show that adapting to the hu-
man’s reward function results in the highest trust in the robot.
1 Introduction
As autonomous technologies become more ubiquitous, the
need to ensure that these technologies behave in a trustwor-
thy manner increases. When working with humans in collab-
oration, these technologies (e.g., autonomous robots, intelli-
gent decision aids, etc.) are being perceived more as team-
mates rather than tools to be used by a human operator. In
such hybrid teams, trust has been identified as a key fac-
tor to facilitate effective and efficient collaboration (Sheri-
dan 2016; Yang, Schemanske, and Searle 2023). To enable
such trust-driven partnerships, it is essential for a robot to
be able to estimate its human partner’s level of trust in real
time. Further, it also needs a way to estimate the human’s
behavior based on her level of trust. Finally, many robotic
decision-making systems use reward maximization to plan
their behaviors. In such cases, it is necessary to ensure that
the “values” of the robot match that of the human. This is
usually accomplished via Inverse Reinforcement Learning
(Ng and Russell 2000), which aims at learning reward func-
tions through observed behaviors.
Although it has been theorized that matching the robot’s
reward function with that of the human in a collaborative
Copyright © 2024, Association for the Advancement of Artificial
Intelligence (www.aaai.org). All rights reserved.
task is good for the team, its effect on trust has not been
studied in detail. Yet there are two reasons to suggest that
such adaptation could be beneficial for trust. First, research
has shown that agent adaptation to humans can enhance per-
formance in a HAT context (Chiou and Lee 2023; Azevedo-
Sa et al. 2020). Second, agent adaptation could be viewed
as the agent being responsive to the human and may, in
turn, increase human trust of the agent (Li et al. 2021). This
study investigates the effect on trust when humans interact
with robots with different interaction strategies. We compare
three types of interaction strategy: (1) the robot does not
align its reward with the human, (2) the robot does not align
its reward with the human, but uses the estimated human re-
ward function for performance assessment, trust estimation,
and behavior prediction, and (3) the robot aligns its reward to
that of the human. We conduct a human-subjects study with
12 participants. Our results indicate that adapting to human
preferences leads to the highest level of trust of the human
on the robot, and leads to a significantly higher number of
agreements with the robot’s recommendations. We use the
term “robot” and “intelligent agent” interchangeably in this
paper.
The rest of the paper is organized as follows: Section 2
gives an overview of related work that our study builds upon.
Section 3 details the human-robot team task and formu-
lates our problem as a trust-aware Markov Decision Process
(trust-aware MDP). Section 4 details the human-subjects ex-
periment. Section 5 discusses major results and their impli-
cations. Finally, section 6 concludes our study and discusses
limitations and future work.
2 Related Work
This work is motivated by two bodies of research, quantita-
tive trust models and value alignment in HRI.
2.1 Trust Models in HRI
Xu and Dudek (2012) proposed a reputation based trust
model to adapt the behavior of robots when trust crossed
a certain threshold. Later, the authors proposed an On-
line Probabilistic Trust Inference Model (OPTIMo) (Xu
and Dudek 2015) which modeled trust through a Dynamic
Bayesian Network. Subsequently, they used this model to
adapt a robot’s behavior depending on the trust level of the
human (Xu and Dudek 2016). Our work differs from this
arXiv:2309.05179v1 [cs.RO] 11 Sep 2023
in the way that we do not use a threshold-based adapta-
tion strategy. Rather, we use an embedded human-behavior
model to predict human behavior with trust and use it di-
rectly in the decision-making process of the robot.
Guo and Yang modeled human trust as a beta distribution
with personalized parameters that are updated during inter-
action depending on the performance of the robot (Guo and
Yang 2021; Yang, Guo, and Schemanske 2023). They pre-
sented simulation results using a reverse-psychology human
behavior model and found that the robot can “manipulate”
human’s trust if an explicit trust-gaining reward is not added
to its reward function (Guo, Shi, and Yang 2021). Bhat et al.
(2022) used this model with a trust-gaining reward term and
demonstrated its usage in a human-subjects study. Our work
is similar to this work in the sense that we use the same
trust estimation model, but we use a different human behav-
ior model, which we call the “bounded-rationality-disuse”
model, which gets rid of this trust manipulation issue as long
as the values of the human and the robot are aligned. Thus,
no trust-gaining reward term is needed in the robot’s reward
function, which makes the reward function much simpler.
2.2 Value Alignment in Human-Robot Teams
The problem of aligning the “values” of the robot to that of
its human teammate has been studied in human-robot team-
ing literature (Hadfield-Menell et al. 2016; Milli et al. 2017;
Fisac et al. 2020; Brown, Schneider, and Niekum 2020;
Yuan et al. 2022). In a majority of these works, techniques
from Inverse Reinforcement Learning (IRL) (Ng and Rus-
sell 2000) are used to estimate a reward function that aligns
with human demonstrations, preferences, or actions.
A bidirectional value alignment problem is studied in
(Yuan et al. 2022). The human knows the true reward func-
tion and behaves accordingly while interacting as a super-
visor to a group of worker robots. The robots try to learn
this true reward function through correctional inputs to their
behavior from the human. The human, on the other hand,
tries to update her belief on the robot’s belief of the true
reward function and inputs corrections to their behavior ac-
cordingly. In our case, there is no true reward function: the
human and the robot have their own reward functions, and
we want to see the effect of aligning/not aligning the robot’s
reward function with that of the human.
A “driver’s test” to verify value alignment between the
human and the robot is provided in Brown, Schneider, and
Niekum (2020). This is especially relevant when the human
and the robot are performing separate tasks in collaboration,
since in this case, it is not enough to match the reward func-
tions of the human and the robot. In our case, since the action
sets for the robot and the human are the same, it is enough to
match their reward functions to guarantee value alignment.
We use a Bayesian framework for IRL (Ramachandran
and Amir 2007) which learns human preferences by main-
taining and updating a distribution over the possible prefer-
ences of the human. The update happens in a Bayesian way
after observing the human’s selected action.
3 Problem Formulation
3.1 Human-Robot Team Task
We designed a scenario in which the human-robot team per-
forms a search for potential threats in a town. The team se-
quentially goes through search sites to look for threats. At
each site, the team is given a probability of threat presence
inside the site via a scan of the site by a drone. The robot
additionally, has some prior information about the probabil-
ity of threat presence at all of the search sites. This prior
information is unknown to the human. After getting the up-
dated probability of threat presence, the robot generates a
recommendation for the human. It can either recommend
that human use or not use an armored robot for protection
from threats. Encountering a threat without protection from
the armored robot will result in injury to the human. On the
other hand, using the armored robot takes extra time since
it takes some time to deploy and move the armored robot to
the search site. The goal of the team is to finish the search
mission as quickly as possible while also maintaining the
soldier’s health level. Thus, a two-fold objective arises with
conflicting sub-goals: To save time you must take risks, and
if you want to avoid risks, you must sacrifice precious mis-
sion time.
3.2 Trust-Aware Markov Decision Process
We model the interaction between the human and the
robot as a trust-aware Markov Decision Process (trust-
aware MDP). A trust-aware MDP is a tuple of the form
(S, A, T, R, H), where Sis a set of states one of which is
the trust of the human in the robot, Ais a finite set of ac-
tions, Tis the transition function giving the transition prob-
abilities from one state to another given an action, Ris a re-
ward function and His an embedded human trust-behavior
model, which gives the probabilities of the human choosing
a certain action given the action chosen by the robot, their
level of trust, etc.
States The level of trust t∈[0,1].
Actions The recommender robot has two choices of ac-
tion: recommend to use or not use the armored robot. These
are represented by ar= 1 and ar= 0 respectively.
Reward Function The rewards for both agents (the hu-
man and the robot) are a weighted sum of the negative cost
of losing health and losing time. The weights for these costs
can be different for the robot and the human. For agent
o∈ {h, r}, the reward function can be written as,
Ro(D, a) = −wo
hh(D, a)−wo
cc(a).(1)
Here, Dis a random variable representing the presence of
threat inside a search site, ais the action chosen by the hu-
man to implement, o∈ {h, r}represents the agent, either
the human hor the robot r.h(D, a)gives the health loss
cost and c(a)gives the time loss cost.
Transition Function The transition function gives the
dynamics of trust as the human interacts with the robot.
We use the model from (Guo and Yang 2021; Bhat et al.
2022) which models trust as a random variable follow-
ing the Beta distribution based on personalized parameters
(α0, β0, ws, wf).
ti∼Beta(αi, βi),
αi=α0+
i
X
j=1
pj,
βi=β0+
i
X
j=1
(1 −pj).
(2)
Where iis the number of interactions completed between the
human and the robot, tiis the current level of trust, and pjis
the realization of the random variable performance (Pj)of
the recommender robot at the jth interaction,
Pj=1,if Rh
j(ar
j)≥Rh
j(1 −ar
j),
0,otherwise. (3)
Here, Rh
j(ar
j)is the reward for the human for choosing the
recommended action (ar
j)at the jth interaction and Rh(1 −
ar
j)is the same for the other action.
Human Trust-Behavior Model A human trust-behavior
model gives the probabilities of a human choosing an ac-
tion, given the robot’s action, their trust level, and other fac-
tors such as the human’s preferences. In our study, we use
the Bounded Rationality Disuse Model as the human trust-
behavior model. This model states that the human chooses
the recommended action with a probability equal to the hu-
man’s current level of trust. If the human chooses to ignore
the recommendation, she will choose an action according to
the bounded rationality model of human behavior. That is,
she will choose an action with a probability that is propor-
tional to the exponential of the expected reward associated
with that action. Mathematically,
P(ah
i=a|ar
i=a) = ti+ (1 −ti)qa,(4)
P(ah
i= 1 −a|ar
i=a) = (1 −ti)(1 −qa).(5)
where qais the probability of choosing action aunder the
bounded rationality model,
qa=exp(κE[Rh
i(a)])
Pa′∈{0,1}exp(κE[Rh
i(a′)]) (6)
3.3 Bayesian Inverse Reinforcement Learning
We use Bayesian IRL to estimate the reward weights of the
human as they interact with the recommender robot. This
is accomplished by maintaining a distribution of the possi-
ble reward weights and updating it using Bayes’ rule after
observing the human’s behavior. More precisely, if bi(w)is
the belief distribution on the reward weights before the ith
interaction, the distribution after the ith interaction is given
by,
bi+1(w)∝P(ah
i=ar
i|ar
i)bi(w),if ah
i=ar
i,
P(ah
i= 1 −ar
i|ar
i)bi(w),otherwise.
(7)
In our formulation, we only learn a distribution over the
health reward weight of the human, wh
h, and assume that the
time reward weight is defined by wh
c:= 1 −wh
h. Further, we
use the mean of the maintained distribution as an estimate
for the human’s health reward weight.
4 Experiment
This section provides details about the testbed used for data
collection and the human-subject experiment. The experi-
ment complied with the American Psychological Associa-
tion code of ethics and was approved by the Institutional
Review Board at the University of Michigan.
4.1 Testbed
We updated some elements of the testbed from (Bhat et al.
2022) for this study, following feedback from participants
in that study. The testbed was developed in the Unreal En-
gine game development platform. We updated the recom-
mendation interface to be more informative and to help the
participants make their own decisions if they choose to do
so. In the updated interface (shown in fig. 1), the partici-
pants are shown the probability of threat presence reported
by the drone, the recommendation given by the intelligent
agent, and an estimate of search time with and without the
armored robot. In order to better separate the threat detection
task from the recommendation task, the participants were
told that a separate entity called an intelligent agent will
given them action recommendations. On the other hand, the
drone’s task is to just scan a site and report the threat level
inside it. The participants were specifically asked to report
their trust level on the intelligent agent.
Figure 1: The recommendation interface
Further, we updated the trust feedback slider to provide
information about the last interaction that the participant
had, in order to help them make an informed decision about
their trust. The updated interface can be seen in fig. 2.
We designed a within-subjects study. Our goal is to com-
pare trust between different interaction strategies. Given the
high variation between trust dynamics between participants
(Bhat et al. 2022), we think that it is better to compare
trust within-subjects than between-subjects. The participants
completed 3 missions in total. In each mission, they inter-
acted with an intelligent agent following one of the 3 inter-
action strategies (detailed in sec. 4.3). In each mission, they
sequentially searched through 40 search sites. The condition
order was counterbalanced using a Latin square.
Figure 2: The trust feedback slider used to get feedback from
the participants after every search site. The mission timer is
paused when the slider is shown to let the participants take
their time in adjusting their trust.
4.2 Participants
We collected data from 12 participants (Age: Mean = 21.92
years, SD =2.36). All participants were students from the
College of Engineering at the University of Michigan.
4.3 Interaction Strategies
We designed three interaction strategies for the intelligent
agent:
•Non-learner: The intelligent agent does not learn the re-
ward weights of the human. It assumes that the human
and the intelligent agent share the same reward weights.
•Non-adaptive learner: The intelligent agent learns per-
sonalized reward weights for each human. It only uses
these learned weights for performance assessment and
human behavior modeling. It still optimizes the MDP
based on its own fixed reward weights.
•Adaptive learner: The intelligent agent learns person-
alized reward weights for each human. It uses them for
performance assessment, human behavior modeling, and
also optimizes the MDP based on these reward weights.
In other words, it updates its own reward function match
the learned reward function.
Although it may look like the non-learner and the non-
adaptive learner both optimize the same reward function,
they actually optimize expected reward under the assumed
human trust-behavior model. Thus, since the non-adaptive-
learner has a better estimate of the human’s preferences, we
postulate that it will have a better estimate of the human’s
trust and behavior, and hence, will show some difference in
its recommendations compared to the non-learner.
4.4 Measures
Pre-experiment Measures Before the beginning of each
mission, we ask the participants to rate their preference be-
tween saving the soldier’s health and saving time by moving
a slider between these two objectives, showing their relative
importance.
In Experiment Measures After each site’s search was
completed, the participants were asked to provide feedback
on their level of trust on the intelligent agent’s recommen-
dations. The interface can be seen in fig. 2. The slider values
were between 0 and 100 with a step of 2 points. The end-
of-mission trust (used in sec 5) is the feedback given by the
participant using this slider after completing the search of
the last search site.
Post-mission Measures We used the following measures
as a post-mission survey that the participants filled out after
every mission.
•Post-mission Trust: This was measured using Muir’s
trust questionnaire (Muir and Moray 1996). It has 9 ques-
tions each with a slider answer range between 0 and 100.
Note that this is separate from the end-of-mission trust,
which is a subjective rating on a single slider by the par-
ticipant after the last search site is completed.
•Post-mission Reliance Intentions: Measured using the
scale developed in (Lyons and Guznov 2019). It has 10
items but 6 items were used herein, each on a 7-point
Likert scale.
5 Results and Discussion
This section provides an overview of our major results and
discusses some reasons behind them and their implications.
Note: All error bars on bar charts are standard errors.
5.1 Trust
We expect that adapting to human preferences will result in
higher levels of trust of the human on the robot.
Figure 3: Mean and standard error (SD) of trust ratings given
by the participants post-mission using Muir’s trust scale
(Muir and Moray 1996)
Fig. 3 shows the post-mission trust rating given using
Muir’s trust scale. Repeated measures ANOVA shows sig-
nificant difference between the trust ratings given to the
three strategies (F(2,22) = 11.962, p < 0.001), with
the highest trust given to the adaptive. Post-hoc analysis
with Bonferroni adjustment shows significant differences
between the non-learner strategy and the adaptive-learner
strategy (p= 0.001) and between the non-adaptive-learner
strategy and the adaptive-learner strategy (p= 0.014).
Figure 4: Average trust reported by the participants across
the interaction period, mean and standard error.
Fig. 4 shows the average trust rating given by the par-
ticipants to the recommendations of the intelligent agent
across their interaction period. Repeated measures ANOVA
shows significant difference between the three strategies
(F(2,22) = 4.968, p = 0.017). Post-hoc analysis with
Bonferroni adjustment reveals significant difference in av-
erage trust rating between the non-learner and the adaptive
learner strategy (p= 0.044) and no significant difference
between the other two pairs.
Figure 5: Trust reported by the participants at the end of the
mission, mean and standard error.
Fig. 5 shows the trust rating given by the participants to
the recommendations of the intelligent agent at the end of
their mission. Repeated measures ANOVA shows signifi-
cant difference between the three strategies (F(2,22) =
7.455, p = 0.003). Post-hoc analysis with Bonferroni ad-
justment reveals significant difference in average trust rat-
ing between the non-learner and the adaptive learner strat-
egy (p= 0.044) and a marginally significant difference be-
tween the non-adaptive-learner and adaptive-learner strate-
gies (p= 0.057). This trend could reach significance with
more data which we are currently working on. The end-of-
Figure 6: Number of agreements between the recommenda-
tion from the robot and the human’s action choice, mean and
standard error.
mission trust rating should be a stable trust rating since the
participants have had enough interactions with the intelli-
gent agent to have a good sense of their trust on it.
Fig. 6 shows the number of agreements between the rec-
ommendation from the intelligent agent and the participant’s
action selection. We expect there to be a positive correla-
tion between the number of agreements and trust reported
by the participants. Repeated measures ANOVA shows sig-
nificant difference between the three strategies (F(2,22) =
13.732, p < 0.001). Post-hoc analysis with Bonferroni ad-
justment reveals significant difference in average trust rat-
ing between the non-learner and the adaptive learner strat-
egy (p= 0.003) and between the non-adaptive-learner and
the adaptive-learner strategy (p= 0.009). The participants
agreed the most with the adaptive-learner’s recommenda-
tions.
6 Conclusions
In this study, we provided a demonstration of the use of
Bayesian IRL coupled with the bounded-rationality-disuse
model of human behavior to learn a human’s preferences in
performing a human-robot team task. We implemented an
adaptive interaction strategy for the robot that learns and op-
timizes a reward function based on these preferences. We
showed the trust and performance improvement when using
such an adaptive interaction strategy compared to two base-
lines.
The results of our study should be seen in light of the
following limitations. First, we provide a demonstration in
the case when there are only two components in the team’s
reward function. Therefore, we only need to learn the hu-
man’s preference for one of the two components and can as-
certain their relative preference between the two objectives.
Our formulation, however, can readily be extended to the
case where there are more than two objectives in the team’s
reward function, with additional computations required to
learn and maintain a distribution over each reward weight.
Second, we used an uninformed uniform prior for the reward
weights of the human. This uniform distribution was, thus,
also used to set the reward weights for the non-adaptive in-
teraction strategies. This simulates a scenario where we do
not have any other way of setting the reward weights for
the robot. Another approach could be to use a data-driven
prior on the reward weight distribution to set these weights.
A similar comparison between the three strategies with such
an informed prior could be a direction for a future study.
Acknowledgments
This work was supported by the Air Force Office of Scien-
tific Research under grant #FA9550-20-1-0406.
References
Azevedo-Sa, H.; Jayaraman, S. K.; Yang, X. J.; Robert, L. P.;
and Tilbury, D. M. 2020. Context-Adaptive Management of
Drivers’ Trust in Automated Vehicles. IEEE Robotics and
Automation Letters, 5(4): 6908–6915.
Bhat, S.; Lyons, J. B.; Shi, C.; and Yang, X. J. 2022. Cluster-
ing Trust Dynamics in a Human-Robot Sequential Decision-
Making Task. IEEE Robotics and Automation Letters, 7(4):
8815–8822.
Brown, D. S.; Schneider, J.; and Niekum, S. 2020. Value
Alignment Verification. CoRR, abs/2012.01557.
Chiou, E. K.; and Lee, J. D. 2023. Trusting Automation:
Designing for Responsivity and Resilience. Human Factors,
65(1): 137–165. PMID: 33906505.
Fisac, J. F.; Gates, M. A.; Hamrick, J. B.; Liu, C.; Hadfield-
Menell, D.; Palaniappan, M.; Malik, D.; Sastry, S. S.; Grif-
fiths, T. L.; and Dragan, A. D. 2020. Pragmatic-Pedagogic
Value Alignment. In Amato, N. M.; Hager, G.; Thomas,
S.; and Torres-Torriti, M., eds., Robotics Research, 49–57.
Cham: Springer International Publishing. ISBN 978-3-030-
28619-4.
Guo, Y.; Shi, C.; and Yang, X. J. 2021. Reverse Psychology
in Trust-Aware Human-Robot Interaction. IEEE Robotics
and Automation Letters, 6(3): 4851–4858.
Guo, Y.; and Yang, X. J. 2021. Modeling and Predicting
Trust Dynamics in Human-Robot Teaming: A Bayesian In-
ference Approach. International Journal of Social Robotics.
Hadfield-Menell, D.; Dragan, A.; Abbeel, P.; and Russell, S.
2016. Cooperative Inverse Reinforcement Learning.
Li, H.; Ni, T.; Agrawal, S.; Jia, F.; Raja, S.; Gui, Y.; Hughes,
D.; Lewis, M.; and Sycara, K. 2021. Individualized Mutual
Adaptation in Human-Agent Teams. IEEE Transactions on
Human-Machine Systems, 51(6): 706–714.
Lyons, J. B.; and Guznov, S. Y. 2019. Individual differences
in human–machine trust: A multi-study look at the perfect
automation schema. Theoretical Issues in Ergonomics Sci-
ence, 20(4): 440–458.
Milli, S.; Hadfield-Menell, D.; Dragan, A.; and Russell, S.
2017. Should Robots be Obedient? ArXiv:1705.09990 [cs].
Muir, B.; and Moray, N. 1996. Trust in automation. Part II.
Experimental studies of trust and human intervention in a
process control simulation. Ergonomics, 39: 429–60.
Ng, A. Y.; and Russell, S. 2000. Algorithms for Inverse Re-
inforcement Learning. In in Proc. 17th International Conf.
on Machine Learning, 663–670. Morgan Kaufmann.
Ramachandran, D.; and Amir, E. 2007. Bayesian Inverse Re-
inforcement Learning. In Proceedings of the 20th Interna-
tional Joint Conference on Artifical Intelligence, IJCAI’07,
2586–2591. San Francisco, CA, USA: Morgan Kaufmann
Publishers Inc.
Sheridan, T. B. 2016. Human–Robot Interaction: Status and
Challenges. Human Factors, 58(4): 525–532. Publisher:
SAGE Publications Inc.
Xu, A.; and Dudek, G. 2012. Trust-driven interactive visual
navigation for autonomous robots. In 2012 IEEE Interna-
tional Conference on Robotics and Automation, 3922–3929.
Xu, A.; and Dudek, G. 2015. OPTIMo: Online Probabilistic
Trust Inference Model for Asymmetric Human-Robot Col-
laborations. In 2015 10th ACM/IEEE International Confer-
ence on Human-Robot Interaction (HRI), 221–228.
Xu, A.; and Dudek, G. 2016. Maintaining efficient collabo-
ration with trust-seeking robots. In 2016 IEEE/RSJ Interna-
tional Conference on Intelligent Robots and Systems (IROS),
3312–3319.
Yang, X. J.; Guo, Y.; and Schemanske, C. 2023. From
Trust to Trust Dynamics: Combining Empirical and Com-
putational Approaches to Model and Predict Trust Dynamics
In Human-Autonomy Interaction. In Duffy, V. G.; Landry,
S. J.; Lee, J. D.; and Stanton, N., eds., Human-Automation
Interaction: Transportation, Automation, Collaboration, &
E-Services, 253–265. Cham: Springer International Publish-
ing.
Yang, X. J.; Schemanske, C.; and Searle, C. 2023. Toward
Quantifying Trust Dynamics: How People Adjust Their
Trust After Moment-to-Moment Interaction With Automa-
tion. Human Factors, 65(5): 862–878.
Yuan, L.; Gao, X.; Zheng, Z.; Edmonds, M.; Wu, Y. N.;
Rossano, F.; Lu, H.; Zhu, Y.; and Zhu, S.-C. 2022. In
situ bidirectional human-robot value alignment. Science
Robotics, 7(68): eabm4183.
