An Artificial Life Environment for Autonomous Virtual Agents

By Toni Conde and Daniel Thalmann* Our approach is based on the multi-sensory integration of the standard theory of neuroscience, where signals of a single object coming from distinct sensory systems are combined. The acquisition steps of signals, filterin

COMPUTER ANIMATION AND VIRTUAL WORLDS

Comp.Anim.Virtual Worlds 2004;15:311–318(DOI:10.1002/cav.34)

******************************************************************************************************An Arti?cial Life Environment for

Autonomous Virtual Agents with

multi-sensorial and multi-perceptive

features

By Toni Conde and Daniel Thalmann*

************************************************************************************

Our approach is based on the multi-sensory integration of the standard theory of

neuroscience,where signals of a single object coming from distinct sensory systems are

combined.The acquisition steps of signals,?ltering,selection and simpli?cation intervening

before proprioception,active and predictive perception are integrated into virtual sensors

and a virtual environment.We will focus on two aspects:1)the assignment problem:

determining which sensory stimuli belong to the same virtual object and (2)the sensory

recoding problem:recoding signals in a common format before combining them.We have

developed three novel methodologies to map the information coming from the virtual

sensors of vision,audition and touch as well as that of the virtual environment in the form of

a ‘cognitive map’.Copyright #2004John Wiley &Sons,Ltd.

KEY WORDS :

arti?cial life;perception;virtual sensors;virtual environment;virtual autonomous agents Introduction

An Autonomous Virtual Agent (AVA)situated in a

Virtual Environment (VE)is equipped with sensors for

vision,audition and touch that inform it of the external

VE and its internal state.An AVA possesses effectors to

let it exert an in?uence on the VE and control architec-

ture to coordinate its perceptions and actions.The

behavior of an AVA is adaptive as long as the control

architecture allows it to maintain its variables in their

validity zone.

Our Arti?cial Life Environment (ALifeE)based on an

original approach inspired by neuroscience equips an

AVA with the main virtual sensors in the form of a small

nervous system.1The control architecture is kept simple

to optimize the management of the AVA’s virtual sensors

and perception.The processes of ?ltering,selection and

simpli?cation are carried out after obtaining the sensorial

information.This approach allows us to achieve some

persistence in the form of a ‘cognitive map’.

The ‘mental processes’of an AVA can be simulated.Behavioral animation includes the techniques applied to make an AVA intelligent and autonomous,to react to its VE and to make decisions based on its perceptive system,its short-term memory and long-term reasoning.Intelligence is the ability to plan and carry out the tasks based on the model of the current state of the VE.Our objective is to permit the AVA to explore un-known VEs and to construct mental structures and models,cognitive maps or plans from this exploration.Once its representation has been created,the knowledge can be communicated to other AVAs.Each AVA perceives objects and other AVAs with the help of its VE,which provides information concerning their nature and posi-tions.The behavioral model decides which action the AVA should take (such as walking or handling an object)and then uses the knowledge.Contributions.In this paper,we present three novel methodologies:*The ?rst technique integrates several virtual sensors in the same multi-sensorial control architecture.*The second technique integrates different perception

types of an AVA coupled with its sensors.

******************************************************************************************************Copyright #2004John Wiley &Sons,

Ltd.*Correspondence to:Daniel Thalmann,Virtual Reality Lab,Swiss Federal Institute of Technology (EPFL),1015Lausanne,Switzerland.E-mail:daniel.thalmann@ep?.ch

By Toni Conde and Daniel Thalmann* Our approach is based on the multi-sensory integration of the standard theory of neuroscience, where signals of a single object coming from distinct sensory systems are combined. The acquisition steps of signals, filterin

*The third technique allows the fusion of multi-

sensorial information in the form of two‘cognitive maps’and‘internal visual memory’before the AVA’s learning and evolving processes.

Document organization:We will begin by examining background research work.Then we will introduce our methodology.Following that,we will present our reali-zation and experimental results.Finally,we will con-clude and suggest possible directions for future work. Virtual Sensors Background

An AVA should be equipped with virtual sensors for vision,audition and touch.These sensors constitute a starting point to implement behavior such as direct vision during a move,handling of objects and respond-ing to sounds or words.

Our ALifeE integrates in an important way the main virtual sensors of an AVA as in the following research: Virtual Vision proposed by,2–4Virtual Audition proposed by5and Virtual Touch proposed by.6After acquiring the information,the basic perceptive part of the AVA is carried out by the Flexible Perception Pipeline approach proposed by.7

Synthetic Vision

Synthetic vision is the main information channel be-tween the VE and the AVA.8,9The AVA perceives its virtual scene from a window and the resulting informa-tion is suf?cient for local navigation.The mixture of image recognition and representation of its VE allows the AVA to react in real time.Noser and al.3have used an octree as an internal …… 此处隐藏：29329字，全部文档内容请下载后查看。喜欢就下载吧 ……