1、Appears in the 2nd International Joint Conference on Autonomous Agents and Multiagent Systems, 2003Modeling Coping Behavior in Virtual Humans:Dont Worry, Be HappyStacy MarsellaUniversity of Southern CaliforniaInformation Sciences Institute4676 Admiralty Way, Marina del Rey, CA 90292marsellaisi.eduJo
2、nathan GratchUniversity of Southern CaliforniaInstitute for Creative Technology13274 Fiji Way, Marina del Rey, CA 90292gratchict.usc.eduABSTRACTThis article builds on insights into how humans cope with emo-tion to guide the design of virtual humans. Although coping isincreasingly viewed in the psych
3、ological literature as having acentral role in human adaptive behavior, it has been largely ig-nored in computational models of emotion. In this paper, weshow how psychological research on the interplay between hu-man emotion, cognition and coping behavior can serve as acentral organizing principle
4、for the behavior of human-likeautonomous agents. We present a detailed domain-independentmodel of coping based on this framework that significantly ex-tends our previous work. We argue that this perspective pro-vides novel insights into realizing adaptive behavior.Keywords: Emotion and personality;
5、human-like qualities1. INTRODUCTIONEmotions play a powerful, central role in our lives. They impactour beliefs, inform our decision-making and in large measureguide how we adapt our behavior to the world around us. Whilemost apparent in moments of great stress, emotions sway eventhe mundane decision
6、s we face in everyday life 1, 2. Emotionsalso infuse our social relationships. Our interactions with eachother are a source of many of our emotions and we have devel-oped both a range of behaviors that communicate emotionalinformation as well as an ability to recognize the emotionalarousal in others
7、. By virtue of their central role and wide influ-ence, emotion provides a means to coordinate the diverse mentaland physical components required to respond to the world in acoherent fashion.We come to the study of emotion with a particular computa-tional perspective. Imagine a computer system that s
8、imulatesNASAs mission control during a crisis, allowing you to prac-tice your crisis management skills. Or consider a system thatallows you to confront your greatest social phobia in the relativesafety of virtual reality. Such social training simulations arepossible through the creation of virtual h
9、umans, software enti-ties that look and act like people, but that live in simulatedgraphical environments 3. To support such dramatic scenarios,virtual humans must model emotion and conveying it in a waythat mirror human emotional behavior. This is essential for be-lievability if an agent looks like
10、 a human, people expect it tobehave like one as well, and will be disturbed by, or misinter-pret, discrepancies from human norms 4. Further, it is essentialfor the learning experience: to support such social training, vir-tual humans must act and make decisions as if they are humansunder stress. The
11、 potential of this virtual human technology isconsiderable. Applications include education and training 5,therapy 6, marketing 7 and entertainment 8. In working on a number of such systems, we have come tothe view that emotion must be treated as more than surface be-havior, but as a central organizi
12、ng construct that can help inte-grate the numerous computational modules that underlie virtualhuman architectures. Virtual humans must act and react in theirsimulated environment, drawing on the disciplines of automatedreasoning and planning. To hold a conversation, they mustexploit the full gamut o
13、f natural language research, from speechrecognition and natural language understanding through naturallanguage generation and speech synthesis. To effectively conveynonverbal behavior, emotion, and personality, they must drawheavily on psychology and communication. It is our view thatemotion plays a
14、 central role in pulling all the agents capabilitiestogether into a believable virtual human. Thus the agents plan-ning, natural language generation, physical behavior, etc. mustbe consistent with its emotional state. Our goal is to model the range of human emotions as well astheir impact on behavio
15、r Although this may seem implausible atfirst, some significant advances in emotion psychology can shedconsiderable light on the design of emotional virtual humans.Specifically, cognitive appraisal theories characterize emotion asthe result of a cognitive appraisal that assesses the relevance ofevent
16、s in terms of their personal significance for an individual.Smith and Lazarus cognitive-motivational-emotive system isone instance of this class of theories 9. Their model viewsemotion as a two-stage control system. Appraisal characterizesthe relationship between a person and their physical and soci
17、alenvironment, referred to as the person-environment relation-ship. Coping recruits resources to repair or maintain this rela-tionship, by motivating actions that change the environment(problem-focused coping), or by motivating changes to its inter-pretation of this relationship (emotion-focused cop
18、ing). Person-ality influences this process through stable biases in how anindividual appraises and copes with events. Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit o
19、r commercial advantage and thatcopies bear this notice and the full citation on the first page. To copyotherwise, or republish, to post on servers or to redistribute to lists,requires prior specific permission and/or a fee.Conference 00, Month 1-2, 2000, City, State.Copyright 2000 ACM 1-58113-000-0/
20、00/0000$5.00.Appears in the 2nd International Joint Conference on Autonomous Agents and Multiagent Systems, 2003The focus of this paper is on a general model of coping, in-cluding its role, along with appraisal, as a central organizingprinciple for human-like autonomous agents. This frameworkhas bee
21、n realized in an implemented system that has been ap-plied to a significant real world virtual human application, theMission Rehearsal Exercise virtual training environment 10.Although considerable research has addressed computationalmodels of emotional appraisal 11-13, there has been almost nowork
22、on coping, at least as generally conceived in the emotionpsychology community. Prior computational work on the impactof emotion on action selection (e.g., 14, 15) can be viewed asproblem-directed coping, however, no work captures the fullrange of human problem-directed and emotion-directed copingbeh
23、avior. Our own prior work provided a preliminary, partialmodel of coping 16. The work we present here significantlybuilds on that work. Specifically, we have extended the model tocover a wide range of both problem and emotion directed strate-gies. Additionally, coping strategies are now modeled with
24、in amore elegant, uniform framework based on select alterations ofappraisal factors. The coping strategy selection process has alsobeen improved in several ways. The potential for a coping strat-egy to be successful is now accounted for in the selection deci-sion. Coping strategies can be combined a
25、nd interleaved, allow-ing more subtle strategies to be realized. Also, we have incorpo-rated a more flexible approach to how coping relates to elicitingevents allowing emotional state to “infect” the agents deci-sion-making in other situations and thus have a more pervasiveimpact, as happens in huma
26、n behavior.2. COGNITION-MOTIVATION-EMOTIONSmith and Lazarus cognitive-motivational-emotive psychologi-cal theory organizes behavior around two basic processes, ap-praisal (which characterizes the persons relationship with theirenvironment), and coping (which suggests strategies for alteringor mainta
27、ining this relationship). Cognition informs both ofthese processes. It informs appraisal by building up mental rep-resentations of how events relate to internal dispositions such asgoals. It informs coping by suggesting and exploring strategiesfor altering or maintaining the person-environment relat
28、ionship.2.1 Appraisal and Appraisal VariablesAppraisal is the process by which a person assesses their overallrelationship with its environment, including not only their cur-rent condition but past events that led to this state as well asfuture prospects. Cognitive appraisal theory argues that peopl
29、epossess many distributed processes for interpreting this relation-ship (e.g., planning, explanation, perception, memory, linguisticprocesses) but that appraisal maps characteristics of these dispa-rate processes into a common set of intermediate terms calledappraisal variables. These variables serv
30、e as an intermediatedescription of the person-environment relationship a commonlanguage of sorts and mediate between stimuli and response(e.g. different responses are organized around how a situation isappraised). Appraisal variables characterize the significance ofevents from the individuals perspe
31、ctive. Events do not havesignificance in of themselves, but only by virtue of their inter-pretation in the context of an individuals beliefs, desires andintention, and past events. For example, the outcome of thelatest presidential election might inspire joy, anger or indiffer-ence, depending on whi
32、ch candidate one desires and ones angertowards an individual may be mitigated by whether one believesthey intended an offensive act. There is broad agreement on the basic set of variables under-lying appraisal (though the complete set and naming of variablesdiffers considerably across theories). In
33、our work, we model thefollowing commonly implicated variables: goal relevance are the consequences of an event relevantto an organisms goals desirability how desirable are the consequences likelihood how likely are the consequences causal attribution who is the causal agent underlying theevent and d
34、o they deserve credit or blame coping potential a measure of an agents ability to reversenegative or maintain positive circumstances.2.2 CopingCoping determines how one responds to the appraised signifi-cance of events. People are motivated to respond to events dif-ferently depending on how they are
35、 appraised 17. For exam-ple, events appraised as undesirable but controllable motivatepeople to develop and execute plans to reverse these circum-stances. On the other hand, events appraised as uncontrollablelead people towards escapism or resignation. Computationalapproaches that model this motivat
36、ional function have largelyfocused on the former response, using emotion or appraisal toguide external action, however psychological theories character-ize coping more broadly. In addition to acting on the environ-ment, termed problem-focused coping, people employ inner-directed strategies for deali
37、ng with strong emotions, termedemotion-focused coping 18. Emotion-focused coping works byaltering ones interpretation of circumstances, for example, bydiscounting a potential threat or abandoning a cherished goal.Indeed, much of what counts as problem-focused coping in thepsychological literature is
38、 really inner-directed in this sense. Forexample, one might form an intention to achieve a desired state and feel better as a consequence without ever acting on theintention. Thus, by performing cognitive acts like planning, onecan improve ones interpretation of circumstances without actu-ally chang
39、ing the physical environment. Beyond organizing coping strategies into these two broadcategories (sometimes researchers add suppression as a thirdseparate category), coping researchers have constructed severaldetailed taxonomies of different techniques people use to cope,often based on peoples subje
40、ctive reports. Table 1 illustrates apartial taxonomy adapted from 19. Coping relies on appraisal to identify significant features ofthe person-environment relationship and to assess the potentialto maintain or overturn these features (coping potential). Basedon these assessments, coping selects amon
41、g competing strategiesto alter the person-environment relationship. For example, ifone feels guilty about causing a traffic accident, one may bemotivated to redress the wrong (problem-focused coping) oralternatively, shift-blame to the other driver (emotion-focusedcoping). Coping typically relies on
42、 cognitive process to actuallyrealize these strategies. So, whereas coping may form the inten-tion to redress the wrong, cognition must still devise a particularplan of attack. The ultimate effect of these strategies is to changea persons interpretation of their relationship with the environ-ment, w
43、hich can lead to a re-appraisal of this relationship. Cop-ing, cognition and appraisal are tightly coupled, interacting andunfolding over time: an agent may “feel” distress for an eventAppears in the 2nd International Joint Conference on Autonomous Agents and Multiagent Systems, 2003(appraisal), whi
44、ch motivates the shifting of blame (coping),which leads to anger (re-appraisal). 3. A COMPUTATIONAL PERSPECTIVEA central tenant in cognitive appraisal theories in general andthe Smith and Lazarus work in particular is that appraisal andcoping center around a persons interpretation of their relation-
45、ship with the environment. This interpretation is constructed bycognitive processes, summarized by appraisal variables andaltered by coping responses. To capture this interpretative proc-ess in computational terms, we have found it most natural tobuild on plan-based causal representations, augmentin
46、g themwith decision-theoretic planning techniques (e.g., 20) and withmethods that explicitly model commitments to beliefs and inten-tions 21, 22. Plan representations provide a concise represen-tation of the causal relationship between events and states, keyfor assessing the relevance of events to a
47、n agents goals and forassessing causal attributions. Plan representations also lie at theheart of a many autonomous agent reasoning techniques (e.g.,planning, explanation, natural language processing). Beyondmodeling causality, attributions of blame or credit involve rea-soning if the causal agent i
48、ntended or foresaw the consequencesof their actions, most naturally represented by explicit represen-tations of beliefs and intentions. As we will see, commitments tobeliefs and intentions also play a key role in implementing cop-ing strategies. The appraisal variables of desirability and likeli-hoo
49、d find natural analogues in the concepts of utility and prob-ability as characterized by decision-theoretic planning methods. In our conceptualization, the agents current interpretation ofits “agent-environment relationship” is reified by the output andintermediate results of those reasoning algorithms that relate theagent to its physical and social environment. We use the termcausal interpretation to refer to this collection of data structuresto emphasize the importance of causal reasoning as well as theinterp
