Feeling crowded yet?: Crowd simulations for VR

Feeling C owded Ye ?: C owd Simula ions o VR Nu ia Pelechano∗ Uni e si a Poli ` ecnica de Ca alunya Jan M. Allbeck† Geo ge Mason Uni e si y ABSTRACT Wi h ad ances in i ual eali y echnology and i s mul iple applica- ions, he need o belie able, imme si e i ual en i onmen s is in- c easing. E en hough cu en compu e g aphics me hods allow us o de elop highly ealis ic i ual wo lds, he main elemen ailing o enhance p esence is au onomous g oups o human inhabi an s. A g ea numbe o c owd simula ion echniques ha e eme ged in he las decade, bu c i ical de ails in he c owd’s mo emen s and appea ance do no mee he s anda ds necessa y o con ince VR pa icipan s ha hey a e p esen in a eal c owd. In his pape , we e iew ecen ad ances in he c ea ion o imme si e i ual c owds and discuss a eas ha equi e u he wo k o u n hese simula ions in o mo e ully imme si e and belie able expe iences. 1 INTRODUCTION Imme sion desc ibes he ex en o which compu e displays a e ca- pable o deli e ing an inclusi e, ex ensi e, su ounding and i id illusion o eali y o he senses o a human pa icipan [27]. In- c eased imme sion can esul in inc eased p esence in he i ual wo ld. Sla e and Wilbu de ine p esence as a s a e o conscious- ness, he (psychological) sense o being in he i ual en i onmen . S udies show ha inc eased p esence in i ual wo lds can yield be e aining ans e om hese g aphical en i onmen s o eal wo ld scena ios [27]. Vi ual c owds ha e po en ial o enhance Vi ual Reali y (VR) expe iences. The way we pe cei e a i ual c owd s ongly depends on he medium and on ou ole. Ou expe ience changes based on whe he we a e a passi e iewe (e.g. mo ies, e en wi h 3D ech- nology) o an ac i e pa icipan (e.g. ideo games). In VR he goal becomes con incing he pa icipan ha he/she is expe iencing some hing eal (i.e. being mo e p esen in he i ual en i onmen han he eal one) and is su ounded by a i ual c owd. Un o u- na ely his is whe e ailu es begin. Pa icipan s’ expec a ion o he i ual c owds a e no me and hei sense o p esence is diminished. The p oblem is ha we as humans obse e each o he e e y day and become accus omed o no mal beha io and a une o ecognizing e en small de ia ions om hese no ms. Any unna u al a i ac s in ou c owd simula ions become magni ied by he i s pe son pe - spec i e and imme si e echnologies o VR. Since we i s explo ed using p esence o e alua e c owd simula- ions [21], a lo o echnologies ha e been de eloped. VR ha dwa e has ad anced and become eadily a ailable as consume le el p od- uc s. C owd simula ions ha e also been g ea ly imp o ed, bu a e no ye o he le el hey need o be o be e ec i e in VR. Resea ch ocuses on: making c owds appea mo e pu pose ul, making cha - ac e s ollow easonable ajec o ies, making anima ions smoo he and mo e na u al, imp o ing ende ing and so on. Ye when im- me sed in VR, c owd simula ions a e s ill ound lacking. Jus as we ix one dis ac ing a i ac , ano he appea s and olls us igh back in o he uncanny alley. We seem o ide an uncanny alley olle coas e . E e y ime we manage o imp o e one ea u e, an issue ∗e-mail: [email p o ec ed] †e-mail: [email p o ec ed] wi h ano he becomes appa en and d as ically educes ealism ye again. So how a a e we om achie ing ou goal o e ec i e c owd simula ions in VR? Tha is he ocus o his pape . We will e alua e cu en c owd simula ion esea ch, we will discuss he p og ess o VR echnology o e he pas eigh yea s, and we will p esen ou hough s on whe e addi ional e o s a e needed o b ing us close o ou goal. 2 CROWD SIMULATION METHODS The e has been a la ge numbe o me hods o simula e c owds o i ual cha ac e s. Mos o he c owd simula ion wo k ocuses on simula ing oo ajec o ies using a bounding olume o he en i e cha ac e o collision a oidance and epulsion o ces. Fo a good su ey on c owd simula ion me hod we e e he eade o ecen books in his ield [13, 19, 30]. Simula ing oo ajec o ies p o ides good esul s in e ms o local mo emen , bu a i ac s appea easily in he anima ions and ee -g ound con ac , specially as he densi ies inc ease and ca e ul ee posi ion and anima ion blending becomes necessa y. In hose si ua ions i is easy o spo oo sliding, ab up changes in eloci y and o so o ien a ion, as well as jumps be ween anima ions. These app oaches o en deal wi h a small da abase o anima ions and ei he limi he mo emen o he cha ac e o he a ailable an- ima ions [15], o include some echniques o in e pola e be ween anima ions and apply ime wa ping o sa is y cons ain s be ween he ee and he en i onmen [20]. Some ecen wo k a emp s o achie e mo e na u al esul s by planning in he oo domain [31, 25] and hen compu ing he igh sequence o anima ions and blending o ollow he esul ing ajec- o y [8]. Ano he aspec o i ual c owds ha equi es a en ion is en- de ing. O en when la ge c owds a e equi ed, i is necessa y o use echniques o speed up ende ing while minimizing loss o i- sual quali y. Bu is ealis ic ende ing eally equi ed o p esence? The e ha e been s udies on isual ealism a ec ing p esence om he poin o iew o he en i onmen unde di e en illumina ion models [32]. F om hose s udies i appea s ha wha ma e s is he dynamics o he illumina ion model, i.e. changing shadows and e- lec ions. The quali y o he illumina ion does no seem o a ec p esence, howe e i esul s in g ea e plausibili y and hus pa ic- ipan s a e mo e likely o espond o i ual e en s as i hey we e eal. This opens he doo o e alua ing he isual appea ance o he c owd. Wo k by Ja abo e al. [12] s udied how c owd mo emen cha ac e is ics and di e en illumina ion models a ec ed he pe - cep ion o c owds. The pu pose o his wo k was o de e mine how much he illumina ion models could be simpli ied o c owd ende - ing wi hou being no iceable. This could also ha e an applica ion o ende ing c owds in IVEs (Imme si e Vi ual En i onmen s). E al- ua ing p esence unde a a ie y o c owd ende ing echniques [7] would help us o de e mine he mos ele an condi ions o use each ype o ende ing and hus o e all sa e esou ces while s ill p o- iding ealis ic isualiza ion o many cha ac e s. Fo example, in Figu e 1 we compa e simple ca oonish looking cha ac e s used in an imme si e i ual c owd expe imen agains a c owd ende ed wi h a mo e human-like appea ance and shadows ha change dy- namically as he cha ac e s mo e. The nex le el o isual appea - © 2016 IEEE. Pe sonal use o his ma e ial is pe mi ed. Pe mission om IEEE mus be ob ained o all o he uses, in any cu en o u u e media, including ep in ing/ epublishing his ma e ial o ad e ising o p omo ional pu poses,c ea ing new collec i e wo ks, o esale o edis ibu ion o se e s o lis s, o euse o any copy igh ed componen o his wo k in o he wo ks. DOI: 10.1109/VHCIE.2016.7563568 ance would be pho o- ealis ic cha ac e s, bu his would mos likely p esen p oblems wi h he Uncanny Valley. Figu e 1: C owds ende ed wi h ca oon like cha ac e s and no shad- ows [22] (le ) o a mo e human-like appea ance and shadows [6] ( igh ). 3 VR TECHNOLOGY When conside ing he po en ial impac s o c owd simula ions on i ual eali y and ice e sa, we mus include bo h so wa e and ha dwa e. Focusing i s on so wa e, he mo e open elease o games engines like Uni y 3D [1] and he Un eal Engine [2] ha e p o ided c owd simula ion esea che s access o echnologies and me hods ha p e iously equi ed g ea e o on hei pa . E en i he ocus o hei esea ch was on s ee ing o s epping beha io s, hey needed o de elop so wa e o na iga ion, ende ing, e c. in o de o show how hei me hods ope a e in a ull c owd simula ion. While no ye ideal and no ye en i ely simplis ic o u ilize, hese game engines p o ide a amewo k ha includes a leas udimen- a y (and in some elemen s mo e ad anced) componen s on which esea che s can examine and de elop hei simula ions. Func ional- i y includes, le el design ools, na iga ion meshes, anima ion play- back and blending, in e se kinema ics, ad anced ende ing ech- niques, and physics engines. Pushes owa d open sou ce so wa e has also lessen he ini ial in es men o ime and e o p e iously needed when beginning c owd simula ion esea ch. Fu he mo e, hese engines p o ide easy access o a ied inpu and ou pu de ices. Resea che s no longe need o be conce ned wi h de eloping de ice d i e s and connec i e so wa e o game con olle s and Head Moun ed Displays (HMDs). These engines p o ide au oma ic connec ions wi h consume le el ha dwa e like wi eless con olle s, acke s (e.g. he Mic oso Kinec ) and he Occulus Ri . This p omo es much g ea e expe imen a ion a lowe in es men cos s. Focusing on VR ha dwa e, i now wo ks and is e en com- mon place [26]. CAVES now ha e high esolu ion displays wi h seamless image alignmen s. The e a e e en a numbe o po able CAVES. HMDs also bene i om high esolu ion as well as wide ields o iew, ligh e weigh , and lowe cos poin s. The e a e now simple moun s o sma phones o become HMDs. Head acke s and whole body acke s, like he Mic oso Kinec , a e common and o many applica ions ha e he equi ed accu acy and ease o use. The a ailabili y o hese de ices makes i much mo e easible o place simula ed c owds in VR and po en ially p o ide new lines o esea ch in pa icipan in e ac ion wi h c owds. De ices such as eye acke s and biome ic senso s (e.g. EEGs and SCRs) ha e also become much mo e pe asi e and a e used o e alua e VR applica ions including he le els o p esence o he pa icipan s [28]. Using hese de ices o aid in he de e mina ion o a subjec ’s le el o p esence in VR wo lds wi h a ying c owd models could be a iable a enue o c owd simula ion e alua ion. 4 STATE OF THE ART IN IMMERSIVE VIRTUAL CROWDS Imme si e i ual c owds could o e a g ea po en ial o e alua e human beha io in di icul o dange ous si ua ions. Vi ual eali y o e s he possibili y o eplica ing expe imen s wi hou exposing he pa icipan o a eal dange ous si ua ion. Fo example, he e has been wo k s udying human beha io s unde a i e eme gency in a public place [28]. This pa icula s udy was ca ied ou wi h a ew i ual agen s wi h sc ip ed beha io s and anima ions. Vi ual c owds in imme si e en i onmen s ha e been p oposed as bo h a alida ion echnique o c owd simula ion om an egocen ic poin o iew, and also as a se up o lea n abou human decision-making in con olled si ua ion [21]. The lea ned da a could be used la e on o enhance c owd simula ion models. In o de o be able o bo h alida e and ex ac ele an da a, i is essen ial o c ea e an en i onmen whe e people eel p esence, and hus eac in a ealis ic way o i ual e en s. In [21] he au ho s p esen a s udy based on p esence li e a u e and ea u es ha appea o en in c owd simula ion models. They discuss how hose ea u es could enhance o b eak p esence. In a la e expe imen [22] using a head moun ed display, hey showed ha ha ing some le el o in- e ac ion wi h he c owds could enhance he eeling o being he e. Fo example, he i ual agen s we e able o see he pa icipan o pe o m collision a oidance, hey could also be pushed by he pa - icipan in case o high densi ies. The pa icipan was imme sed in a i ual cock ail pa y whe e he c owd would walk a ound and s op a ables wi h ood and d inks. Agen s would au onomously decide o s op a ound ables, u n o ace o he agen s, and play idle beha io s ha included hand ges u es imi a ing being engaged in con e sa ion. All o hese elemen s combined wi h he backg ound noise o pla es and con e sa ion helped c ea e a s ong sense o p esence despi e he ca oonis appea ance o he cha ac e s. CAVE sys ems ha e also been used o imme se a pa icipan in i ual c owd o he pu pose o e alua ing new c owd simula ion me hods [3]. In his wo k he goal was o compa e di e en c owd simula ion me hods. The pa icipan was s anding in a loca ion wai ing o a pe son and could obse e he c owds om an ego- cen ic poin o iew o e alua e hei collision a oidance me hods. Ques ionnai es we e used a e wa d o ga he in o ma ion abou he pa icipan s opinion ega ding li e-likeness o he c owd, spa ial awa eness and com o in he CAVE. Bo h he wo k by Pelechano e al. [22] and Ahn e al. [3] ound ha b eaks in p esence o low com o appea when he e is no col- lision a oidance (i.e. cha ac e s mo ing h ough he pa icipan ) and when he me hod p oduces shaking a i ac s. The wo k by Ky iakou e al. [14] demons a ed ha showing col- lision a oidance be ween he use and he i ual agen s wi h some basic in e ac ion be ween hem could inc ease he sense o p es- ence. This kind o in e ac ion also helped o make he i ual en i- onmen appea mo e ealis ic and li elike. In hei s udies hey also obse ed ha including in e ac ion wi h he i ual c owd could en- hance p esence. The ypes o in e ac ion used o hei wo k we e limi ed o ha ing cha ac e s wa ing o e bally salu ing he pa ic- ipan . The wo k by B uneau e al. [9] s udied he pa icipan s’ beha io when walking agains g oups o i ual agen s. The goal was o de e mine wha elemen s o he c owd ge people o decide o walk h ough o a ound a g oup o people. The elemen s o he g oup ha hey es ed whe e based on ela i e mo ion, size, densi y o appea ance. P oxemics ha e also been e alua ed wi h small g oups o i ual cha ac e s walking owa ds he pa icipan [17]. Those expe imen s ound ha physiological a ousal would inc ease as he cha ac e s would app oach he pa icipan and also as he numbe o i ual humans would inc ease om 1 o 4. These las wo pape s had some in e es ing, i no su p ising, e- sul s which we e ha in ce ain si ua ions he appea ance o he i - ual humans did no seem o ha e much e ec on he pa icipan ’s esponse. Fo example he i s one [9] was e alua ing he pa ic- ipan s’ decisions o walking h ough o a ound a g oup o i ual humans. I u ned ou ha whe he he i ual humans we e en- de ed as o dina y people o zombies had no signi ican di e ences in he decisions made by pa icipan s. And simila ly, in [17] hey ound no signi ican di e ences in e ms o physiological a ousal when being app oached by i ual people o cylinde s. This opens ques ions ega ding he ex end o which ealis ic isual appea ance is necessa y and why. IVE p o ides a pla o m o s udy c owd beha io unde con- olled condi ions [18]. In his pape hey discuss and e alua e he use o VR o ca y ou con olled expe imen s o ob ain da a abou human beha io in e ms o na iga ion. The long e m goal would be o be able o use VR as an aid in he p ocess o de eloping mo e accu a e mic oscopic models. The pape aises in e es ing ques- ions such as: a e ajec o ies in VR simila o hose in eal li e? o , is he e any bias induced by he use o locomo ion in e aces? The answe s o hese ques ions a e s ill un esol ed. Fo ins ance, pa o he wo k in his ield is ocusing on enhancing p esence in an a emp o ha e he pa icipan s beha ing as in eal li e. Bu how close a e eal esponses o i ual e en s om ha ing pa icipan s pe o ming ealis ic local mo emen ? Gi en ha mo emen i sel is cons ained in mos VR applica ions, (i.e. cables in HMD o size o he CAVEs) can we ully us pa icipan mo emen ? In o he wo ds, is p esence enough o u n VR in o an accu a e pla o m o cap u e human locomo ion? The e has been a la ge amoun o wo k ying o ind co ela ions be ween human walking in s a ic i ual en i onmen s and in eal se ups ha exac ly ma ch he dimensions o he i ual one. La ely some esea che s ha e been mo ing hese s udies o dynamic en i onmen s popula ed wi h i ual humans o ind he simila i ies be ween he eal and he i ual wo ld [24]. Despi e ecen ad ances in VR and he a ailabili y o engines such as Un eal [2] and Uni y [1] ha acili a e he ask o c ea - ing i ual en i onmen s, he e is s ill a la ge amoun o manual wo k equi ed o popula e hose en i onmen s in a belie able man- ne . The pape by Spanlang e al. [29] p o ides a good o e iew o he di icul ies in c ea ing a i ual expe imen dealing wi h embod- imen . Embodimen equi es he eplacemen o he pa icipan ’s body wi h a i ual cha ac e seen om a i s pe son pe spec i e. To be belie able, he body ep esen a ion needs o ollow he mo e- men s o he pa icipan . This implies ha ing a leas one i ual human, wi h eal- ime mo ion cap u e, cha ac e anima ion, and e- alis ic ende ing. Ha ing a ew belie able i ual cha ac e s al eady equi es a la ge amoun o wo k, bu he p oblems escala e as he numbe o cha ac e s needed inc eases, since i is necessa y o deal no only wi h he simula ion o each cha ac e bu also wi h he in- e ac ions be ween hem. Mos o he ime hese i ual cha ac e s ollow s ic ly sc ip ed beha io s and hei appa en in e ac ion simply ollows he Wiz- a d o Oz me apho . Anima ions a e mo ion cap u ed and ca e ully cleaned and adjus ed o he inal scena io. In such expe imen s, he e is li le oom o au onomous beha io , and he i ual hu- mans a e like he ac o s in a mo ie wi h li le oom o in e ac ion wi h he pa icipan . When we deal wi h imme si e c owds, he job o edi ing scena ios wi h such le el o de ail becomes un easible, and hus mo e esea ch is needed o mo e p esence expe imen a- ion in o he c owd simula ion ield. 5 SOWHAT’SMISSING? In [26] Sla e discusses g and challenges in i ual en i onmen s, a eas ha need u he esea ch. One opic highligh ed is acking and he need o be e body acking o e wide a eas wi h high accu acy and low la ency. Such ad ancemen s in acking could posi i ely impac c owd simula ion esea ch on wo on s. One, i would enable pa icipan s imme sed in a i ual c owd o mo e mo e na u ally and mo e eely. I could also enable be e da a collec ion o inpu o c owd simula ions and o e alua e hem. Sla e also discusses plausibili y illusion. Pa icipan s ca y ex- pec a ions abou hei en i onmen and he people in i om he eal wo ld in o he i ual wo lds. This calls o mo e ealis ic hu- man beha io s including he beha io s o c owds and hei in e ac- ions wi h he en i onmen . Said in e ac ions equi e a high le el o domain knowledge, he c ea ion o which is edious and ime consuming. Recen esea ch has explo ed au oma ing he c ea ion o elemen s o domain knowledge [23, 5], bu addi ional e o is needed. The plausibili y illusion also calls o esea che s o a oid he Uncanny Valley. When pa icipan expec a ions a e no me , p es- ence can be b oken. When pa icipan expec a ions a e se oo high by one aspec o he simula ion, lowe quali y in ano he aspec can b eak p esence. Along hese lines, he e a e a eas o ac i e esea ch ha need con inued imp o emen o c owd simula ions in VR o be success ul. We need be e anima ion and blending ools, be e ende ing o la ge c owds, be e ing s epping and na iga ion algo- i hms (pa icula ly in clu e ed en i onmen s). In ac anima ion bugs and gli ches end o be e y no iceable in VR and e y dis- ac ing. We also need o calib a e he speed o cha ac e s when placed in VR. We ha e obse ed ha hey end o be pe cei ed as mo ing mo e quickly han hey should. Acco ding o Sla e , ano he challenge o i ual en i onmen s is hap ics. While he e a e an a ay o hap ic de ices a ailable, none can eplica e all, o pe haps e en any, o he ouches and eels we expe ience daily. Sla e poin s ou ha esea ch in he a ea o hap ics has o con end wi h and is limi ed by he laws o physics, making i pa icula ly challenging. A lack o ealis ic hap ic eed- back is p oblema ic when imme sing pa icipan s in c owds. The lack o ac ile eedback om e en ela i ely mino collisions wi h indi iduals in he c owd may g ea ly dec ease p esence. Absence o he o ces and p essu es no mally expe ienced in dense c owds would g ea ly dec ease a pa icipan ’s sense o p esence. Resea ch on how o eplica e he eeling o being imme sed in a c owd is s ill needed. Do ad ances in neu oscience hold he solu ions? Ano he challenge is enabling people nai e o p og amming o c ea e and modi y en i onmen s. Cu en so wa e amewo ks like he game engines discussed in Sec ion 3 ha e begun add essing his challenge. They p o ide ools o impo ing models, designing le - els, blending anima ions, designing simple AIs, e c. They do no , howe e , p o ide ools o he c ea ion and con ol o i ual hu- man popula ions. While we ha e explo ed au ho ing he beha io s o popula ions om a limi ed numbe o easy o au ho ac ion ypes (See igu e 2) [19, 16, 4], addi ional esea ch is needed. Figu e 2: Simple GUIs o au ho ing eac i e, s ochas ic, and need- based ac ions (le ) and Mic oso Ou look calenda s o au ho ing scheduled ac ions ( igh ). Cummings e al., who s udied imme sion and p esence h ough a me a-analysis o p e ious esea ch [11], discussed di e en ypes o p esence. They indica e ha social p esence would be inc eased h ough be e communica ion and social ac o s. Ce ainly his is an elemen o c owd simula ions ha could use addi ional esea ch. Cha ac e s in c owds end o be exp essionless and la gely non e- ac i e. While hey may espond o a oid collisions and o a ew sc ip ed e en s, hey end o appea a he li eless. They possess limi ed communica ion models and ew social ea u es. E en he addi ion o models o a en ion would p o ide mo e li elike qual- i ies. Cha ac e s should look whe e hey a e going and a in e - es ing objec s and e en s in he wo ld. Indi iduals in c owds ha mee he eye o he pa icipan om ime o ime could help o in- c ease social p esence. Cha ac e s should eac o each o he and o he pa icipan in a na u al manne . Figu e 3 shows ha e en in a a he p imi i e c owd simula ion and VR se up, pa icipan s can be posi i ely in luenced by e en unin en ional cha ac e ac ions. To inc ease social p esence, cha ac e s should exhibi su p ise, cu ios- i y, and e en ange . They should exhibi di e en pe sonali ies and a i udes. They should appea ...human. Figu e 3: A pa icipan in a VR expe imen on p esence eac ing o a pe cei ed wa e om a i ual cha ac e by s a ing o wa e back (mo emen o le hand). In e ms o spa ial p esence hey ound acking le el, s e e- oscopy, and ield-o - iew o be mo e impac ul han image qual- i y, esolu ion, and sound. T acking p o ides pa icipan s wi h a sense o sel -loca ion, be e na iga ion, and addi ional ac ion pos- sibili ies. S e eoscopy p o ides g ea e spa ial p esence h ough in- c eased dep h pe cep ion and ce ainly wide ields-o - iew aid in sel loca ion. They ound ha he abili y o na iga e a ound he en- i onmen is key o inc easing spa ial p esence. Na iga ion is mo e di icul in c owds, especially dense c owds. Will including c owds in VR inhe en ly dec ease le els o spa ial p esences o will pa ic- ipan s ha e lowe expec a ions o na iga ion in c owds and ha e hei le els o p esence emain high? Ano he elemen o c owd simula ions ha has no ecei ed much a en ion is sound. Adding ealis ic, localized sound o he cha ac e s in a simula ion may imp o e he pa icipan ’s le el o p esence. They would be less s a led by cha ac e s suddenly ap- pea ing in hei ield-o - iew. The cha ac e s hemsel es could e- ac o noises in he en i onmen , making hem appea mo e li elike. Finally, elemen s o s o y and game play wi hin c owd simula- ions may inc ease imme sion and p esence. I pa icipan s a e en- h alled by a s o y o ac i i y wi hin he i ual en i onmen , hey may no ice ewe anomalies and be less impac ed by hem. Pe haps elemen s o s o y and game play need o be add essed in c owd simula ion esea ch. 6 CONCLUSION VR can bene i om ad ances in c owd simula ion echniques and c owd simula ion esea ch can bene i om ad ances in VR ech- nology. Popula ions o i ual humans can make imme si e i - ual wo lds mo e ealis ic, in e es ing, and compelling. The c owd simula ion communi y con inues o explo e ways o e alua ing and alida ing hei echniques. VR is an e e mo e iable enue o imme sing pa icipan s in a c owd and examining hei beha io al eac ions and o e all le els o p esence. Howe e , as ou lined in his pape c owd simula ions mus s ill imp o e o mee he s an- da ds needed o achie e high le els o p esence. Since VR p esen s an ideal pla o m o s udy human beha io in c owds, i is hus impo an o ocus ou esea ch e o s in de e - mining how we can achie e ealis ic imme si e c owds. One way ha we ound e y aluable o make p og ess is o ob ain mo e in o ma ion om he pa icipan s a e he expe imen s. So o in- s ance, once all he esul s om ques ionnai es and physiological esponse ha e been ga he ed, we can ask he pa icipan s o gi e us gene al commen s abou he expe imen . Wha was he mos com- pelling? Wha did hey ind was missing? Wha b oke he eeling o in e ac ing wi h a eal c owd? Thei answe s can help he e- sea che s o hink ou side he box by no icing wha he a e age pe - son may be expe iencing as opposed o wha he esea che eels is missing. I also p o ides ele an in o ma ion ega ding wha he a e age pe son may no ca e abou ha he esea che s do. Fo example, when playing ideo games, people ge so imme sed and exci ed abou he game ha hey olun a ily decide o igno e he a i ac s (i.e. low [10]). So he playe does no s op chasing he en- emy jus because he/she obse ed a bad blending, an ab up change in di ec ion, o some oo sliding. Could pa icipan s in a p esence en i onmen wi h an imme si e c owd also adap o he ci cum- s ances and decide o igno e ce ain a i ac s? Maybe as long as he i ual humans appea eal by alking o hem, wa ing o showing emo ions, he pa icipan would ea hem as eal people. Mo e esea ch is needed. 7 ACKNOWLEDGMENTS This wo k has been pa ially unded by he Spanish Minis y o Science and Inno a ion and FEDER unde g an TIN2014-52211- C2-1-R. REFERENCES [1] Uni y Game Engine 5. h p://uni y3d.com/es/, 2016. [2] Un eal Engine 4. h ps://www.un ealengine.com/wha -is-un eal- engine-4, 2016. [3] J. Ahn, N. Wang, D. Thalmann, and R. Boulic. Wi hin-c owd imme - si e e alua ion o collision a oidance beha io s. In P oceedings o he 11 h ACM SIGGRAPH In e na ional Con e ence on Vi ual-Reali y Con inuum and I s Applica ions in Indus y, VRCAI ’12, pages 231– 238, New Yo k, NY, USA, 2012. ACM. [4] J. M. Allbeck. CAROSA: A ool o au ho ing NPCs. In Mo ion in Games, pages 182–193. Sp inge Be lin Heidelbe g, 2010. [5] T. Balin and J. Allbeck. Au oma ed Gene a ion o Plausible Agen Objec In e ac ions. In W.-P. B inkman, J. B oekens, and D. Heylen, edi o s, In elligen Vi ual Agen s, olume 9238 o Lec u e No es in Compu e Science, pages 295–309. Sp inge In e na ional Publishing, 2015. [6] A. Beacco, C. Anduja , N. Pelechano, and B. Spanlang. E icien en- de ing o anima ed cha ac e s h ough op imized pe -join impos o s. Compu e Anima ion and Vi ual Wo lds, 23(1):33–47, 2012. [7] A. Beacco, N. Pelechano, and C. Andja . A su ey o eal- ime c owd ende ing. Compu e G aphics Fo um, pages n/a–n/a, 2015. [8] A. Beacco, N. Pelechano, M. Kapadia, and N. I. Badle . Foo s ep pa ame e ized mo ion blending using ba ycen ic coo dina es. Com- pu e s & G aphics, 47:105 – 112, 2015. [9] J. B uneau, A.-H. Oli ie , and J. Pe e. Going h ough, going a ound: a s udy on indi idual a oidance o g oups. Visualiza ion and Com- pu e G aphics, IEEE T ansac ions on, 21(4):520–528, 2015. [10] M. Csikszen mihalyi. Flow: The Psychology o Op imal Expe ience. Ha pe Pe ennial Mode n Classics, 2008. [11] J. J. Cummings and J. N. Bailenson. How imme si e is enough? a me a-analysis o he e ec o imme si e echnology on use p esence. Media Psychology, pages 1–38, 2015. [12] A. Ja abo, T. V. Eyck, V. Sunds ed , K. Bala, D. Gu ie ez, and C. O’Sulli an. C owd ligh : E alua ing he pe cei ed ideli y o illu- mina ed dynamic scenes. Compu e G aphics Fo um, 31(2p 3):565– 574, 2012. [13] M. Kapadia, N. Pelechano, J. Allbeck, and N. Badle . Vi ual c owds: S eps owa d beha io al ealism. Syn hesis Lec u es on Visual Com- pu ing: Compu e G aphics, Anima ion, Compu a ional Pho og aphy, and Imaging, 7(4):1–270, 2015. [14] M. Ky iakou, X. Pan, and Y. Ch ysan hou. In e ac ion wi h i ual agen s. compa ison o he pa icipan s’ expe ience be ween an i and a semi-i sys em. In Vi ual Reali y (VR), 2015 IEEE, pages 217–218. IEEE, 2015. [15] M. Lau and J. J. Ku ne . P ecompu ed sea ch ees: Planning o in e ac i e goal-d i en anima ion. In P oceedings o he 2006 ACM SIGGRAPH/Eu og aphics Symposium on Compu e Anima ion, SCA ’06, pages 299–308, Ai e-la-Ville, Swi ze land, Swi ze land, 2006. Eu og aphics Associa ion. [16] W. Li and J. M. Allbeck. Popula ions wi h pu pose. In Mo ion in Games, pages 132–143. Sp inge Be lin Heidelbe g, 2011. [17] J. Llobe a, B. Spanlang, G. Ru ini, and M. Sla e . P oxemics wi h mul iple dynamic cha ac e s in an imme si e i ual en i onmen . ACM T ans. Appl. Pe cep ., 8(1):3:1–3:12, No . 2010. [18] A.-H. Oli ie , J. B uneau, G. Ci io, and J. Pe ´ e. A i ual eali y pla o m o s udy c owd beha io s. T anspo a ion Resea ch P oce- dia, 2:114–122, 2014. [19] N. Pelechano, J. M. Allbeck, and N. I. Badle . Vi ual c owds: Me h- ods, simula ion, and con ol. Syn hesis Lec u es on Compu e G aph- ics and Anima ion, 3(1):1–176, 2008. [20] N. Pelechano, B. Spanlang, and A. Beacco. A a a locomo ion in c owd simula ion. In e na ional Jou nal o Vi ual Reali y, 10(1):13, 2011. [21] N. Pelechano, C. S ocke , J. Allbeck, and N. Badle . Feeling c owded? explo ing p esence in i ual c owds. In P oceedings o PRESENCE 2007. The 10 h annual In e na ional Wo kshop on P esence, pages 373–376, 2007. [22] N. Pelechano, C. S ocke , J. Allbeck, and N. Badle . Being a pa o he c owd: Towa ds alida ing c owds using p esence. In P oceedings o he 7 h In e na ional Join Con e ence on Au onomous Agen s and Mul iagen Sys ems - Volume 1, AAMAS ’08, pages 136–142, Rich- land, SC, 2008. In e na ional Founda ion o Au onomous Agen s and Mul iagen Sys ems. [23] C. Pelkey and J. M. Allbeck. Popula ing Vi ual Seman ic En i on- men s. Compu e Anima ion and Vi ual Wo lds, 24(3):405–414, May 2014. [24] J. Pe ine , A.-H. Oli ie , and J. Pe ´ e. Walk wi h me: In e ac ions in emo ional walking si ua ions, a pilo s udy. In P oceedings o he ACM Symposium on Applied Pe cep ion, SAP ’13, pages 59–66, New Yo k, NY, USA, 2013. ACM. [25] S. Singh, M. Kapadia, G. Reinman, and P. Falou sos. Foo s ep na i- ga ion o dynamic c owds. Compu e Anima ion and Vi ual Wo lds, 22(2-3):151–158, 2011. [26] M. Sla e . G and challenges in i ual en i onmen s. F on ie s in Robo ics and AI, 1(3), 2014. [27] M. Sla e and S. Wilbu . A amewo k o imme si e i ual en i on- men s ( i e): Specula ions on he ole o p esence in i ual en i on- men s. P esence: Teleope a o s and i ual en i onmen s, 6(6):603– 616, 1997. [28] B. Spanlang, T. F ¨ ohlich, V. F. Descalzo, A. An ley, and M. Sla e . The making o a p esence expe imen : Responses o i ual i e. In Annual In e na ional Wo kshop on P esence, pages 303–307, 2007. [29] B. Spanlang, J.-M. No mand, D. Bo land, K. Kil eni, E. Giannopou- los, A. Pomes, M. Gonzalez-F anco, D. P ez Ma cos, J. A oyo Pala- cios, X. N. Muncunill, and M. Sla e . How o build an embodimen lab: Achie ing body ep esen a ion illusions in i ual eali y. F on- ie s in Robo ics and AI, 1(9), 2014. [30] D. Thalmann and S. R. Musse. C owd simula ion. 2013. [31] B. J. H. an Bas en, S. A. S ¨ u el, and A. Egges. A hyb id in e po- la ion scheme o oo p in -d i en walking syn hesis. In P oceedings o G aphics In e ace 2011, GI ’11, pages 9–16, School o Compu e Science, Uni e si y o Wa e loo, Wa e loo, On a io, Canada, 2011. Canadian Human-Compu e Communica ions Socie y. [32] I. Yu, J. Mo ensen, P. Khanna, B. Spanlang, and M. Sla e . Vi- sual ealism enhances ealis ic esponse in an imme si e i ual en i onmen -pa 2. Compu e G aphics and Applica ions, IEEE, 32(6):36–45, 2012.