Moving Virtual Agents Forward in Space and Time Gabriel F. Silva Virtual Humans Lab

2025-05-02 0 0 475.76KB 10 页 10玖币

侵权投诉

Moving Virtual Agents Forward in Space and Time

Gabriel F. Silva

Virtual Humans Lab

PUCRS

Porto Alegre, Brazil

gabriel.fonseca94@edu.pucrs.br

Carlos G. Johansson

Virtual Humans Lab

PUCRS

Porto Alegre, Brazil

carlos@ufcspa.edu.br

Paulo Knob

Virtual Humans Lab

PUCRS

Porto Alegre, Brazil

paulo.knob@edu.pucrs.br

Soraia R. Musse

Virtual Humans Lab

PUCRS

Porto Alegre, Brazil

soraia.musse@pucrs.br

Douglas A. Schlatter

Virtual Humans Lab

PUCRS

Porto Alegre, Brazil

douglas.schlatter01@edu.pucrs.br

Abstract—This article proposes an adaptation from the model

of Bianco for fast-forwarding agents in crowd simulation, which

enables us to accurately fast forward agents in time. Besides being

able to jump from one position to another, agents are able to stay

inside their track, it means, the new position is calculated taking

into account the original global path the agent would follow,

if not being fast-forwarded. Obstacles and other agents around

are also taken into account when calculating the new position. In

addition, we included a personality aspect on agents, which affect

their behaviors and, also, be taken into account when jumping

to a future time and space. We conducted some experiments to

validate our model, which shows that it was able to indeed fast

forward agents from a position to another, in a coherent time,

sticking to a given global path while avoiding collisions. Finally,

we present a use case, showing that our method can ﬁt inside a

”Fog of War” system.

Index Terms—crowd simulation, virtual agents, fast forward-

ing

I. INTRODUCTION

1Since the pioneer work proposed by Thalmann and

Musse [1], many other methods were proposed for crowd

simulation, each one with a signiﬁcant contribution. There are

methods that deal with crowds from a microscopic point of

view [2], [3], as well methods that deal with a macroscopic

point of view [4], [5] and, even, methods which combine

both microscopic and macroscopic simulation strategies [6].

Others explored high dense crowds [2], [7], heterogeneous

behaviors [8], navigation control [9] and personality traits for

agents [10], [11].

Despite the great number of methods proposed for the most

varied range of subjects concerning crowd simulation, only

very few of them tackled the problem of fast-forwarding a

simulation or, in other words, instantly jumping agents from

a position A in time X to another position B in time Y, while

maintaining a coherent path and with a minimum error. Such

a feature is especially useful if performance is taken into

account. For example, if a simulation is fast-forwarded from

frame 200 to frame 300, this interval of 100 frames does not

1Draft version made for arXiv: https://arxiv.org/

need to be simulated, relieving computational resources. In

their work, Bianco et al. [12] proposed a method to estimate

the future position of agents in crowd simulations. A Pedes-

trian Dead Reckoning (PDR) method is used, evaluating the

prior positions, velocities, and goals of each agent. The ﬁnal

positions are then estimated based on a global environment

complexity factor, that aims to impact the fast-forwarding

process, as well as the interaction among agents. Their model

was extended by Bianco et al. [13] to allow the inclusion of

events (e.g. adding an obstacle) during the jumping period.

In games, movement estimation of agents in virtual envi-

ronments is a common topic of research, especially regarding

computer-controlled units in Real-Time Strategy (RTS) games.

Due to ”Fog of War” systems (i.e. vision restricted to only the

player units), that estimation must be implemented considering

a partially observable environment. Hagelback and Johanso

[14] use potential ﬁelds to control a bot with imperfect

information (i.e. affected by the Fog of War) to explore

and navigate the environment. Cho et al. [15] explore the

effect of Fog of War system in predictions made by machine

learning algorithms. Another common approach includes giv-

ing perfect information about the environment to computer-

controlled units, allowing the movement estimation to deal

with obstacles and terrain deformation hidden by the Fog

of War. This approach can be combined with a fast forward

method, reducing the computational cost of units navigating

through the fog.

Although interesting, the method of Bianco et al. [12]

has space to improve. The global environment complexity

factor is too general, which means that the more complex the

environment, the more chances to have a high error rate and,

thus, agents jumping to positions far away from where they

should be. Therefore, in this work, we aim to adapt the model

of Bianco et al. [12] to make it more accurate. So, given an

initial and ﬁnal time/frame, each agent is able to teleport from

its current position A to a future position B, following their

own path and time constraints. Thus, we are not only able

to teleport agents, but to avoid collision with obstacles and

arXiv:2210.03073v1 [cs.MA] 6 Oct 2022

maintain a coherent path based on the agent current position

and its goal. Also, our agents can have personality traits, which

impact their behaviors and are, also, taken into account in the

fast-forwarding method. Finally, as a use case, we integrate our

adapted model with a Fog of War system, placing occluded

agents in a ”suspended” state until the estimated position is

reached or a visible area obstruct its path.

This paper is divided as follows: Section II presents the

related work regarding position estimation of agents in crowd

simulations and real-world pedestrians, along with methods

regarding crowd simulation optimizations during movement

prediction. Section III presents the methodology used for

integrating a path planning algorithm and personality traits of

agents within the fast-forwarding model, as well as how we

include personality traits to our agents. Section IV presents the

results achieved in the integration process and a use case of our

proposed method in a fog of war system. Section V presents

the ﬁnal considerations and future work of our method.

II. RELATED WORK

Due to the increasing demands of simulated environments,

such as a higher number of agents and larger scenarios, several

methods have been presented to reduce the computational

cost of agent movement, collision detection, and collision

avoidance on crowd simulations. Pettre et al. [16] presented

a method where a level of simulation (LOS), similar to a

level of detail (LOD), is given to different sections of the

navigation mesh based on the camera viewpoint. Agents closer

to the camera are updated with higher frequency. Farther

away and occluded agents are given lower priority. Osborne

and Dickinson [17] presented a similar method for grouping

agents using a hierarchical level of detail (LOD). A group

can be deﬁned as a set of agents of the same type, goal,

and nearby positions. The LOD of a group is deﬁned based

on its distance to the camera. Guy et al. [18] proposed a

parallel method for local collision avoidance using the concept

of velocity obstacles. Each agent takes into consideration the

current velocity of nearby agents to create a set of cones, each

containing the possible directions that will cause a collision.

After that, the direction of each agent is adjusted seeking a

point not included inside the cones, which avoids collision

between them.

Different methods and navigation techniques to estimate

the position of real-life pedestrians, which were later adapted

for virtual agents, are presented in the literature. Beaure-

gard and Haas [19] used a pedestrian dead reckoning (PDR)

method combined with acceleration sensors to estimate indoor

positions. Taia et al. [20] combines a PDR technique with

A* path planning algorithm [21] to provide a more precise

approximation of pedestrian routes. Yi et al. [22] proposed

a method for estimating the travel time for pedestrians in

crowded scenes. Their model deﬁnes a group of regions of

interest (ROI) for pairings of both sources and destinations,

calculating the trafﬁc ﬂow and densities for each region.

Environmental elements and stationary persons are taken into

account as obstacles. Abnormal behaviors, such as wandering

pedestrians, can be identiﬁed based on the deviation from their

estimated travel time.

Bianco et al. [12] presented a method for estimating the fu-

ture individual position of virtual agents in crowd simulations.

Their work uses a PDR method to deﬁne the prior positions

of each agent, taking into consideration their goal and speed

at the frame where the jump occurs. Future positions are

adjusted based on a global environment complexity factor and

interaction with other agents. This model was later extended

by Bianco et al. [13] to allow the inclusion of events during

the time jump period. Events are deﬁned as changes in the

environment (i.e. adding, moving of removing obstacles and

goals). Also, a metric for comparing crowds was presented,

taking into consideration the local densities of uniform regions

in the environment to deﬁne a global error estimation.

Differing from methods that focus on collision detection

optimization or levels of detail, we propose a method that

integrates the Pedestrian Dead Reckoning (PDR) position

estimation with a global path planning algorithm, allowing us

to simulate virtual agents that are aware of obstacles in the

environment, including during the fast-forwarding. We also

included personality traits in agents that affect their behavior

during the simulation, aiming to evaluate their impact on the

path planning algorithm and the fast forward process.

III. PROPOSED MODEL

Fig. 1. Comparison between the model presented by Bianco et al [15] (left)

and our proposed method (right). An agent is placed on the bottom-left corner

of the environment, aiming to reach its goal (triangle). The square represents

an obstacle obstructing the agent’s path. The position of the agent after the

fast forward is represented by a plus sign. Our method introduces a global

path planning algorithm, that adjusts the agent’s path based on obstructing

obstacles, in order to fast forward the agent position in time and space.

The main goal of this work is to adapt the model of Bianco

et al. [12] to make it more accurate, regarding the impact of

obstacles and motion trajectories. To do so, we propose to

remove the global environment complexity factor, that was

computed taking into account the global free space to move

in the environment. Instead, we control the jumping using a

navigation method. In other words, such a navigation method

would deliver the global path each agent should follow to reach

its goal. When the agent needs to jump in time, its new position

is calculated taking such a path into account. Also, we include

a personality factor in our agents, which should inﬂuence their

behavior and be taken into account when fast-forwarding the

simulation. Section III-A presents our method to fast forward

the simulation, while Section III-B presents our method to

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：

MovingVirtualAgentsForwardinSpaceandTimeGabrielF.SilvaVirtualHumansLabPUCRSPortoAlegre,Brazilgabriel.fonseca94@edu.pucrs.brCarlosG.JohanssonVirtualHumansLabPUCRSPortoAlegre,Brazilcarlos@ufcspa.edu.brPauloKnobVirtualHumansLabPUCRSPortoAlegre,Brazilpaulo.knob@edu.pucrs.brSoraiaR.MusseVirtualHumansLabP...

展开>> 收起<<

Moving Virtual Agents Forward in Space and Time Gabriel F. Silva Virtual Humans Lab.pdf

共10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！

Moving Virtual Agents Forward in Space and Time Gabriel F. Silva Virtual Humans Lab

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: