Proactive Opinion-Driven Robot Navigation around Human Movers Charlotte Cathcart Mar ıa Santos Shinkyu Park and Naomi Ehrich Leonard Abstract We propose analyze and experimentally verify

2025-05-02 0 0 6.64MB 7 页 10玖币

侵权投诉

Proactive Opinion-Driven Robot Navigation around Human Movers

Charlotte Cathcart, Mar´

ıa Santos, Shinkyu Park, and Naomi Ehrich Leonard

Abstract— We propose, analyze, and experimentally verify

a new proactive approach for robot social navigation driven

by the robot’s “opinion” for which way and by how much

to pass human movers crossing its path. The robot forms

an opinion over time according to nonlinear dynamics that

depend on the robot’s observations of human movers and its

level of attention to these social cues. For these dynamics, it

is guaranteed that when the robot’s attention is greater than

a critical value, deadlock in decision making is broken, and

the robot rapidly forms a strong opinion, passing each human

mover even if the robot has no bias nor evidence for which way

to pass. We enable proactive rapid and reliable social navigation

by having the robot grow its attention across the critical

value when a human mover approaches. With human-robot

experiments we demonstrate the ﬂexibility of our approach and

validate our analytical results on deadlock-breaking. We also

show that a single design parameter can tune the trade-off

between efﬁciency and reliability in human-robot passing. The

new approach has the additional advantage that it does not rely

on a predictive model of human behavior.

I. INTRODUCTION

Autonomous mobile robots are increasingly being used

for tasks in settings such as warehouses and open public

spaces where they will encounter human movers. In order

to accomplish their tasks in these settings, the robots need

to reliably and gracefully navigate around human movers. In

this paper, we propose, analyze, and experimentally verify

a new approach for the social navigation of a mobile robot.

Fig. 1 shows experimental results of a mobile robot navigat-

ing around two human movers using the new approach.

We build on the nonlinear opinion dynamics model pre-

sented in [1] and propose an approach that allows a robot

to rapidly form an opinion that represents the strength of its

preference for which direction—left or right—it will use to

pass each human mover crossing its path. This opinion, in

turn, drives the robot’s motion, modifying its nominal path

to reliably pass the human. A key to the opinion dynamics

is that when the robot’s attention to social cues grows above

a critical value, the neutral opinion to stay the course is

destabilized and the robot rapidly forms a strong and stable

opinion for moving in one of the two passing directions.

Our approach is therefore to design dynamics for the robot’s

Cathcart, Santos, and Leonard are with the Department of Mechanical and

Aerospace Engineering, Princeton University, Princeton, NJ 08544, USA.

{cathcart, maria.santos, naomi}@princeton.edu. Park is

with the Electrical and Computer Engineering program, King Abdullah

University of Science and Technology, Thuwal 23955, Saudi Arabia.

shinkyu.park@kaust.edu.sa.

This research has been supported by ONR grant N00014-19-1-2556,

funding from KAUST, and Princeton University through the generosity of

Lydia and William Addy ’82.

Study #14788 has been approved by the International Review Board

(IRB) of Princeton University.

(a) (b)

Fig. 1. A robot using opinion-driven navigation to pass two humans. (a)

Time-lapse of the experimental trial. (b) The full trajectories of the robot

(red line) and two humans (blue and green lines) with temporal markers.

attention that drive it above this critical value when the

robot senses a human mover approaching its path. The active

control of attention yields a rapid and reliable passing motion

in response to an approaching human mover; this renders our

approach “proactive” rather than merely “reactive.”

Once the robot passes a human, its opinion with respect to

that human is no longer relevant; the opinion quickly returns

to its neutral value, allowing the robot to continue towards

its destination. Likewise, the robot’s attention also goes to

zero, making the robot ready for new potential conﬂicts with

other movers. Figs. 1 and 2 provide experimental results of

the robot navigating different encounters when traveling to a

goal destination that is diagonally across an open space with

two humans moving and pausing in a variety of scenarios.

Opinion dynamics are used to enable decision making in

multi-agent systems in a range of tasks [2]–[4]. In the nonlin-

ear opinion dynamics of [1], an agent’s opinion is inﬂuenced

by the opinions of others when its attention exceeds a critical

level. At this point the agents are guaranteed to form strong

opinions (e.g., to agree on or coordinate among options),

hence avoiding indecision, i.e., deadlock in their decision

making. In the robot social navigation problem, we leverage

the deadlock breaking guarantees of the coupled attention-

opinion dynamics to ensure that, when necessary to avoid an

approaching human mover, the robot will rapidly select and

move in one of the two passing directions even if there is

no indication from the human or the environment that one

direction is better than the other, or if the robot’s bias for

one direction or the other, if it has one, conﬂicts with the

human’s chosen passing direction.

Of relevance to our work is the literature on robot social

navigation (see recent survey articles [5]–[9] and references

therein), where a common theme is in investigating the de-

sign of navigation algorithms for autonomous robots to safely

and comfortably interact with the humans they encounter.

arXiv:2210.01642v4 [cs.RO] 11 Sep 2023

(a) (b) (c) (d) (e)

Fig. 2. Multiple experimental trials with two humans and a robot using the new approach. The top row shows the complete trajectories of the robot (red

line) and humans (green and blue lines) over the course of a trial as the robot moves toward its goal (red star). Each trajectory is marked with an arrow

indicating the mover’s direction. The bottom row shows the robot’s opinion zr(teal line) and attention ur(orange line) over the course of the trial above

it. Temporal markers (dots) are shown along spatial trajectories, opinion, and attention. See Section IV-A for parameters used.

Earlier work [10] in modeling human navigation behavior

proposes a model based on the observation that the motion

of pedestrians is subject to social forces. More recent works

[11], [12] incorporate social cues into the social force model

and the improved models are used to design robot naviga-

tion algorithms. The work of [13] proposes a constrained

optimization approach to design a navigation algorithm that

penalizes the robot when its behavior violates conventions

observed in the human’s navigation. In [9], a reactive control

policy is used to follow and maintain the passing sides

observed by passing humans through social momentum. Ref-

erences such as [14]–[16] explain learning-based approaches

that leverage the recent advancement in deep reinforcement

learning to train mobile robots through multiple trial-and-

error processes to safely navigate in human-populated areas.

Another important line of research in the social navigation

literature is data-driven learning approaches that infer human

navigation models from their demonstration data, and use the

models to predict human motions and to design robot motion

planners. The work of [17] leverages Bayesian learning to

construct a motion model and personality characteristics of

pedestrians, and use predicted pedestrian trajectories from

the model for socially-aware robot navigation. Inverse Re-

inforcement Learning (IRL)-based approaches, for instance

[16], [18], [19], take human demonstration data to estimate

a utility function used in human navigation tasks, and use it

to generate robot trajectories that imitate the demonstrated

human motions. In particular, a recent relevant work [20]

studies the effect of human-robot communication in social

navigation and proposes an IRL-based robot planning frame-

work to generate communication actions that maximize the

robot’s transparency and efﬁciency.

Our work is distinct in that 1) it is proactive rather than

reactive, 2) it does not require constructing a predictive

model of human navigation as in IRL-based approaches,

rather it only needs the robot to observe the position and

moving direction of the human, and 3) our robot navigation

model is analytically tractable so that we can establish a

guarantee on deadlock-free decision making in the robot-

human navigation. This contrasts with the reinforcement

learning approaches, which are in general difﬁcult to analyze,

and existing reactive approaches, such as social force models,

which do not provide the same deadlock-free guarantee.

In Section II, we introduce the nonlinear opinion dynam-

ics and propose a new model for robot navigation in a

human-robot navigation setting. In Section III, using tools

from nonlinear dynamical systems theory, we discuss how

the model ensures rapid deadlock-free robot navigation. To

demonstrate and test the ﬂexibility of our approach, we

carry out experiments with two human participants and a

mobile robot in a range of scenarios, which we report on in

Section IV-A. We examine and validate the effectiveness of

rapid deadlock-free navigation with further experiments in

Section IV-B. We conclude with a discussion in Section V.

II. NONLINEAR OPINION DYNAMICS

IN SOCIAL NAVIGATION

We study a robot navigation problem where a robot

approaches and passes human movers while traveling to its

destination (see examples in Figs. 1 and 2). In this context,

we want to enable the robot to repeatedly overcome human

movers in a rapid and reliable fashion. We are also interested

in tackling challenging scenarios such as the human-corridor

passing problem [21]–[23] that may result in deadlock if,

for example, both the robot and the human have conﬂicting

passing biases. In these situations, a key objective is to ensure

that the robot moves reliably around the human regardless

of the human’s awareness of the robot. It is also desirable

that the robot moves efﬁciently around the human. However,

reliability and efﬁciency are in tension: giving the human a

lot of space may create reliably successful but inefﬁcient

passing whereas giving the human only a little space is

efﬁcient but creates less reliably successful passing.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ProactiveOpinion-DrivenRobotNavigationaroundHumanMoversCharlotteCathcart,Mar´ıaSantos,ShinkyuPark,andNaomiEhrichLeonardAbstract—Wepropose,analyze,andexperimentallyverifyanewproactiveapproachforrobotsocialnavigationdrivenbytherobot’s“opinion”forwhichwayandbyhowmuchtopasshumanmoverscrossingitspath.The...

展开>> 收起<<

Proactive Opinion-Driven Robot Navigation around Human Movers Charlotte Cathcart Mar ıa Santos Shinkyu Park and Naomi Ehrich Leonard Abstract We propose analyze and experimentally verify.pdf

共7页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Proactive Opinion-Driven Robot Navigation around Human Movers Charlotte Cathcart Mar ıa Santos Shinkyu Park and Naomi Ehrich Leonard Abstract We propose analyze and experimentally verify

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: