社会意识自主车辆导航的群体情绪分类.pdf

poPoq

105

9页

0次

2021-05-02

50墨值下载

The Socially Invisible Robot

Navigation in the Social World using Robot Entitativity

Aniket Bera

, Tanmay Randhavane

, Emily Kubin

, Austin Wang

, Kurt Gray

, and Dinesh Manocha

Abstract— We present a real-time, data-driven algorithm to

enhance the social-invisibility of robots within crowds. Our

approach is based on prior psychological research, which

reveals that people notice and–importantly–react negatively to

groups of social actors when they have high entitativity, moving

in a tight group with similar appearances and trajectories. In

order to evaluate that behavior, we performed a user study

to develop navigational algorithms that minimize entitativity.

This study establishes mapping between emotional reactions

and multi-robot trajectories and appearances, and further

generalizes the ﬁnding across various environmental conditions.

We demonstrate the applicability of our entitativity modeling

for trajectory computation for active surveillance and dynamic

intervention in simulated robot-human interaction scenarios.

Our approach empirically shows that various levels of entitative

robots can be used to both avoid and inﬂuence pedestrians

while not eliciting strong emotional reactions, giving multi-robot

systems socially-invisibility.

I. INTRODUCTION

As robots have become more common in social environ-

ments, people’s expectations of their social skills have

increased. People often want robots to be more socially

visible–more salient social agents within group contexts [17].

This social visibility includes being more capable of drawing

the attention of humans and evoking powerful emotions [22].

Cases of social visibility include tasks in which robots must

work collaboratively with humans. However, not all contexts

require socially visible robots. There are situations in which

robots are not used to collaborate with people but instead

used to monitor them. In these cases, it may be better for

robots to be socially invisible.

Social invisibility refers to the ability of agents to escape

the attention of other people. For example, psychological

research reveals that African Americans often go unnoticed

in social environments[11], especially reactions related to

threat. Evolution has attuned the human brain to respond

rapidly to threatening stimuli, thus the less a person–or

a robot–induces negative emotion, the less likely it is to

be noticed within a social milieu. The social invisibility

conferred by not inducing emotion is especially important in

surveillance contexts in which robots are expected to move

seamlessly among people without being noticed. Noticing

surveillance robots not only makes people hide their behav-

ior, but the negative emotions that prompt detection may

also induce reactance [9], which may lead to people to lash

out and harm the robots or even other people [12] Research

reveals a number of ways of decreasing negative emotional

Authors from the Department of Computer Science, University of North

Carolina at Chapel Hill, USA

Authors from the Department of Psychology and Neuroscience, Univer-

sity of North Carolina at Chapel Hill, USA

Fig. 1: Multi-robot systems (robots marked by blue trajec-

tories) are used among crowds for surveillance and mon-

itoring. Our novel navigation algorithm takes into account

various levels of physical and social constraints and use them

for: (a) Active surveillance including monitoring crowds (red

trajectories) while moving through them with no collisions;

(b) Dynamic intervention where the robots try to inﬂuence

the crowd behavior and movements and make the pedestrians

avoid the area marked by a yellow overlay. The dashed

red line indicates the predicted pedestrian trajectories if the

robots did not attempt to dynamically intervene.

reactions towards social agents [10], but one element may

be especially important for multi-robot systems: entitativity

[13], “groupiness”) is tied to three main elements, uniformity

of appearance, common movement, and proximity to one

another. The more agents look and move the same, and the

closer agents are to each other, the more entitative a group

seems, which is why a marching military platoon seems more

grouplike than people milling around a shopping mall.

The threatening nature of groups means that the more entita-

tive (or grouplike) a collection of agents seem, the greater the

emotional reaction they induce and the greater their social

visibility. As maximizing the social invisibility of collections

of agents requires minimizing perceptions of threat, it is im-

portant for multi-robot systems to minimize their entitativity.

In other words, if multi-robots systems are to move through

groups without eliciting negative reactions [16], they must

seem more like individuals and less like a cohesive and

coordinated group.

arXiv:1805.05543v2 [cs.RO] 18 Jul 2018

Fig. 2: Our method takes a live or streaming crowd video as an input. We extract the initial set of pedestrian trajectories

using an online pedestrian tracker. Based on the level of social invisibility we want to achieve, we compute motion model

parameters of the robot group using a data-driven entitativity mapping (which we compute based on a user-study(Section

IV)).

Main Results: We present a novel, real-time planning

algorithm that seeks to optimize entitativity within pedestrian

environments in order to increase socially-invisible navi-

gation (by minimizing negative emotional reactions). First,

we conduct a user study to empirically tie trajectories of

multi-robot systems to emotional reactions, revealing that–as

predicted–more entitative robots are seen as more unnerving.

Second, we generalize these results across a number of

different environmental conditions (like lighting). Third, we

extract the trajectory of each pedestrian from the video and

use Bayesian learning algorithms to compute their motion

model. Using entitativity features of groups of robots and

the pedestrians, we perform long-term path prediction for

the pedestrians. To determine these entitativity features we

establish a data-driven entitativity mapping (EDM) between

the group robot motion and entitativity measure from an

elaborate web-based perception user study that compares the

participants’ emotional reactions towards simulated videos

of multiple robots. Speciﬁcally, highly entitative collections

of robots are reported as unnerving and uncomfortable. The

results of our mapping are well supported by psychology

literature on entitativity [34].

We highlight the beneﬁts of our data-driven metric for use

of multiple robots for crowd surveillance and active interfer-

ence. We attempt to provide maximally efﬁcient navigation

and result in maximum social invisibility. In order to pursue

different sets of scenarios and applications, we highlight the

performance of our work in multiple surveillance scenarios

based on the level of increasing social interaction between

the robots and the humans.

Our approach has the following beneﬁts:

1. Entitativity Computation: Our algorithm accurately pre-

dicts emotional reactions (entitativity) of pedestrians towards

robots in groups.

2. Robust computation: Our algorithm is robust and can

account for noise in pedestrian trajectories, extracted from

videos.

3. Fast and Accurate: Our algorithm involves no pre-

computation and evaluates the entitativity behaviors at in-

teractive rates.

The rest of the paper is organized as follows. In Section 2,

we review the related work in the ﬁeld of psychology and

behavior modeling. In Section 3, we give a background on

quantifying entitativity and introduce our notation. In Section

4, we present our interactive algorithm, which computes

the perceived group entitativity from trajectories extracted

from video. In Section 5, we describe our user study on the

perception of multiple simulated robots with varying degrees

of entitativity.

II. RELATED WORK

Human beings are inherently social creatures, making inter-

acting with and perceiving others an important part of the

human experience. Complex interactions within brain regions

work harmoniously to navigate the social landscape [36].

Interesting patterns emerge when attempting to understand

how humans view groups of people.

A. Psychological Perspectives on Group Dynamics

A long-standing tenet of social psychology is that people’s

behaviors hinge upon their group context. Importantly, the

impact of social dynamics is highly inﬂuenced by group

contexts [38]–often for the worse. Decades of psychological

research reveals that people interact more negatively with

groups than with individuals [34], expressing more hostility

towards and feeling more threatened by a group than an

individual [16]. Such reactions to groups have real world

implications, especially when onlookers have the ability to

act violently. At the heart of these anti-social actions are

negative emotional reactions, which can be directed at any

social agent, whether human or robot [19]. Most often, these

emotions are unease [8], threat [19], and fear [30].

B. Human-Aware Robot Navigation

Many approaches have been applied towards the navigation

of socially-aware robots [31], [7], [3], [15], [25], [29], [18],