{"id":52119,"date":"2019-11-01T00:00:00","date_gmt":"2019-10-31T21:00:00","guid":{"rendered":"https:\/\/prohoster.info\/blog\/blog_prohoster\/kak-ustroen-gibridnyj-igrovoj-ii-i-v-chyom-ego-preimushhestva"},"modified":"2020-02-18T13:59:47","modified_gmt":"2020-02-18T10:59:47","slug":"kak-ustroen-gibridnyj-igrovoj-ii-i-v-chyom-ego-preimushhestva","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/kak-ustroen-gibridnyj-igrovoj-ii-i-v-chyom-ego-preimushhestva","title":{"rendered":"Jak funguje hybridn\u00ed hern\u00ed AI a jak\u00e9 m\u00e1 v\u00fdhody","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

V pokra\u010dov\u00e1n\u00ed d\u0159\u00edve zm\u00edn\u011bn\u00e9ho t\u00e9matu na na\u0161em blogu hern\u00edho um\u011bl\u00e9ho inteligence<\/a><\/noindex> se pod\u00edv\u00e1me na to, jak je k n\u011bmu aplikovateln\u00e9 strojov\u00e9 u\u010den\u00ed a v jak\u00e9 podob\u011b. O sv\u00fdch zku\u0161enostech a na jejich z\u00e1klad\u011b p\u0159ijat\u00fdch rozhodnut\u00edch sd\u00edlel<\/a><\/noindex> expert na ot\u00e1zky AI v Apex Game Tools Jakob Rasmussen.<\/p>\n

\"Jak
\n
\nV posledn\u00edch letech se hodn\u011b hovo\u0159\u00ed o tom, \u017ee strojov\u00e9 u\u010den\u00ed radik\u00e1ln\u011b zm\u011bn\u00ed hern\u00ed pr\u016fmysl, nebo\u0165 tato technologie se u\u017e stala pr\u016flomovou v mnoha dal\u0161\u00edch digit\u00e1ln\u00edch aplikac\u00edch. Ale nem\u011bli bychom zapom\u00ednat, \u017ee hry jsou mnohem slo\u017eit\u011bj\u0161\u00ed ne\u017e simul\u00e1tor \u0159\u00edzen\u00ed automobilu, \u0159\u00edzen\u00ed dron\u016f nebo algoritmy rozpozn\u00e1v\u00e1n\u00ed obli\u010dej\u016f na obr\u00e1zc\u00edch. <\/a><\/noindex><\/p>\n

Zat\u00edmco v hern\u00edm pr\u016fmyslu st\u00e1le p\u0159etrv\u00e1v\u00e1 pou\u017e\u00edv\u00e1n\u00ed tradi\u010dn\u00edch metod AI, jako jsou metody kone\u010dn\u00fdch automat\u016f, stromy chov\u00e1n\u00ed a \u2014 v posledn\u00ed dob\u011b st\u00e1le \u010dast\u011bji \u2014 AI zalo\u017een\u00e9 na u\u017eitku. Takov\u00e9 AI se tak\u00e9 naz\u00fdv\u00e1 design-based (um\u011bl\u00e1 inteligence zalo\u017een\u00e1 na designu) nebo expertn\u00edmi syst\u00e9my. Ale st\u00e1v\u00e1 se \u010d\u00edm d\u00e1l jasn\u011bj\u0161\u00edm \u2014 a to p\u0159edev\u0161\u00edm hr\u00e1\u010d\u016fm \u2014 \u017ee tyto syst\u00e9my jsou st\u00e1le m\u00e9n\u011b vhodn\u00e9 pro vytv\u00e1\u0159en\u00ed opravdu pokro\u010dil\u00fdch protivn\u00edk\u016f, kte\u0159\u00ed by byli schopni imitovat chov\u00e1n\u00ed hr\u00e1\u010d\u016f. To se zvl\u00e1\u0161t\u011b t\u00fdk\u00e1 kreativn\u00edch \u0159e\u0161en\u00ed. Lze to vysv\u011btlit t\u00edm, \u017ee v\u00fdvoj\u00e1\u0159i um\u011bl\u00e9 inteligence nejsou schopni vz\u00edt v \u00favahu v\u0161echny mo\u017en\u00e9 taktiky a strategie chov\u00e1n\u00ed a \u00fasp\u011b\u0161n\u011b je implementovat v tradi\u010dn\u00edch AI syst\u00e9mech. Pro hr\u00e1\u010de to \u010dasto znamen\u00e1, \u017ee je nudn\u00e9 a p\u0159edv\u00eddateln\u00e9 hr\u00e1t proti protivn\u00edkovi, jeho\u017e linie chov\u00e1n\u00ed je snadno zapamatovateln\u00e1.<\/p>\n

K tomuto v\u00fdsledku vede mnoho d\u016fvod\u016f, ale jedn\u00edm z hlavn\u00edch je neschopnost AI se u\u010dit. Proto p\u0159i vytv\u00e1\u0159en\u00ed um\u011bl\u00e9 inteligence protivn\u00edka p\u0159irozen\u011b p\u0159ich\u00e1z\u00ed na mysl p\u0159echod na strojov\u00e9 u\u010den\u00ed, kter\u00e9 se osv\u011bd\u010dilo v mnoha jin\u00fdch aplikac\u00edch. Ale existuje n\u011bkolik nuanc\u00ed, kter\u00e9 stoj\u00ed za to zv\u00e1\u017eit. Tak hern\u00ed AI by m\u011bla b\u00fdt schopn\u00e1 se p\u0159izp\u016fsobit jak\u00e9koli situaci a vyu\u017e\u00edvat v\u00fdhody, kter\u00e9 nab\u00eddne, a tak\u00e9 se adaptovat na r\u016fzn\u00e9 hern\u00ed styly soupe\u0159\u016f \u2014 \u017eiv\u00fdch hr\u00e1\u010d\u016f a dal\u0161\u00edch AI. <\/p>\n

<\/p>\n

Jak\u00e1 je aktu\u00e1ln\u00ed situace<\/h3>\n

\nBritsk\u00e1 spole\u010dnost DeepMind, kter\u00e1 se zab\u00fdv\u00e1 v\u00fdvojem um\u011bl\u00e9 inteligence, ned\u00e1vno uk\u00e1zala, jak se m\u016f\u017ee AI sama u\u010dit hr\u00e1t hry, osvojovat si jejich pravidla a nach\u00e1zet zp\u016fsoby, jak proj\u00edt hrou nebo v n\u00ed zv\u00edt\u011bzit \u2014 zat\u00edm v\u0161ak pouze na p\u0159\u00edkladu jednoduch\u00fdch her, jako jsou ran\u00e9 hry Atari<\/a><\/noindex> \u2014 nap\u0159\u00edklad \u0161achy a japonsk\u00e1 logick\u00e1 hra go. V\u00fdsledky z\u00edskan\u00e9 pro n\u011b ukazuj\u00ed, \u017ee um\u011bl\u00e1 inteligence je schopna si vytvo\u0159it adekv\u00e1tn\u00ed hodnocen\u00ed prob\u00edhaj\u00edc\u00edho d\u011bn\u00ed na hrac\u00ed plo\u0161e. Pokud jde o adaptaci AI na r\u016fzn\u00e9 hern\u00ed styly protivn\u00edka, dosavadn\u00ed v\u00fdsledky nejsou zat\u00edm tak p\u016fsobiv\u00e9.<\/p>\n

V dne\u0161n\u00ed dob\u011b ji\u017e neuronov\u00e9 s\u00edt\u011b dok\u00e1\u017eou rozpozn\u00e1vat obr\u00e1zky a \u0159\u00eddit automobily. Tyto funkce v\u0161ak lze realizovat tak\u00e9 pomoc\u00ed relativn\u011b jednoduch\u00fdch architektur, i kdy\u017e se v d\u016fsledku uk\u00e1\u017eou b\u00fdt pom\u011brn\u011b hlubok\u00e9 a objemn\u00e9. Nap\u0159\u00edklad AI pro rozpozn\u00e1v\u00e1n\u00ed obr\u00e1zk\u016f na Facebooku m\u00e1 hloubku p\u0159ibli\u017en\u011b 100 vrstev, co\u017e ji \u010din\u00ed podobnou biologick\u00e9mu mozku \u2014 z hlediska po\u010dtu a slo\u017eitosti vz\u00e1jemn\u00fdch vztah\u016f mezi neurony, kter\u00e9 tvo\u0159\u00ed jednu velkou s\u00ed\u0165. <\/p>\n

<\/p>\n

Hern\u00ed AI<\/h3>\n

\nCo se t\u00fd\u010de pou\u017eit\u00ed strojov\u00e9ho u\u010den\u00ed v hern\u00edm pr\u016fmyslu, existuje \u0159ada omezen\u00ed, kter\u00e1 br\u00e1n\u00ed jeho vyu\u017eit\u00ed v t\u00e9to oblasti. Pat\u0159\u00ed sem syst\u00e9mov\u00e9 po\u017eadavky, zejm\u00e9na ty, kter\u00e9 se vztahuj\u00ed k CPU, je\u017e ur\u010duj\u00ed schopnost po\u010d\u00edta\u010de zpracov\u00e1vat slo\u017eitou hern\u00ed strukturu a jeho vhodnost pro hern\u00ed vypr\u00e1v\u011bn\u00ed a hratelnost. <\/p>\n

Tak\u017ee se ukazuje, \u017ee v mnoha hr\u00e1ch nen\u00ed mo\u017en\u00e9 realizovat slo\u017eit\u00e9 syst\u00e9my um\u011bl\u00e9 inteligence, proto\u017ee chyb\u00ed pot\u0159ebn\u00fd hardware a zejm\u00e9na klastrov\u00e9 servery, kter\u00e9 existuj\u00ed nap\u0159\u00edklad pro s\u00edt\u011b rozpozn\u00e1v\u00e1n\u00ed obraz\u016f na Facebooku. N\u011bkdy mus\u00ed dokonce fungovat n\u011bkolik AI najednou \u2013 nejen na po\u010d\u00edta\u010d\u00edch, ale tak\u00e9 na mobiln\u00edch za\u0159\u00edzen\u00edch a jin\u00fdch m\u00e9n\u011b v\u00fdkonn\u00fdch platform\u00e1ch. To v\u0161e omezuje velikost a slo\u017eitost architektury strojov\u00e9ho u\u010den\u00ed, proto\u017ee ve\u0161ker\u00e9 v\u00fdpo\u010dty mus\u00ed b\u00fdt prov\u00e1d\u011bny s \u010dasem sn\u00edmku kolem 1 nebo 2 milisekund. Samoz\u0159ejm\u011b je mo\u017en\u00e9 vyu\u017e\u00edt r\u016fzn\u00e9 technologie optimalizace a rozd\u011blen\u00ed z\u00e1t\u011b\u017ee mezi sn\u00edmky, ale p\u0159esto se z t\u011bchto omezen\u00ed nelze zcela vymanit.<\/p>\n

Slo\u017eitost hry m\u016f\u017ee zp\u016fsobit v\u00e1\u017en\u00e9 probl\u00e9my pro AI. Nap\u0159\u00edklad v hr\u00e1ch jako StarCraft II je hern\u00ed mechanika mnohem slo\u017eit\u011bj\u0161\u00ed ne\u017e v hr\u00e1ch Atari. Proto nelze o\u010dek\u00e1vat, \u017ee p\u0159i dan\u00e9 sn\u00edmkov\u00e9 frekvenci a zn\u00e1m\u00fdch syst\u00e9mov\u00fdch po\u017eadavc\u00edch se strojov\u00e9 u\u010den\u00ed nutn\u011b sezn\u00e1m\u00ed s cel\u00fdm stavem hry a bude schopno s n\u00edm interagovat. Jak hr\u00e1\u010d \u010dasto spol\u00e9h\u00e1 na intuici v ran\u00fdch f\u00e1z\u00edch hry, tak se AI mus\u00ed nau\u010dit z\u00e1kladn\u00ed zpracov\u00e1n\u00ed stavu hry, aby jednodu\u0161eji pokra\u010dovalo ve h\u0159e. Nap\u0159\u00edklad v jednom z posledn\u00edch. API pro Starcraft II<\/a><\/noindex> na map\u00e1ch se zobrazuj\u00ed pouze ty informace, kter\u00e9 v\u00fdvoj\u00e1\u0159i pova\u017eovali za d\u016fle\u017eit\u00e9: v jednom p\u0159\u00edpad\u011b AI pou\u017e\u00edvalo zmen\u0161enou verzi cel\u00e9 mapy, v druh\u00e9m \u2013 podobn\u011b jako hr\u00e1\u010d, mohlo posouvat kameru, a pak jeho vn\u00edm\u00e1n\u00ed bylo omezeno pouze na informace na obrazovce.<\/p>\n

\"Jak
\nVizualizace hry AI AlphaStar proti hr\u00e1\u010di v StarCraft II: na screenshotu jsou zobrazeny \"syrov\u00e9\" zdrojov\u00e9 pozorov\u00e1n\u00ed, aktivita neuronov\u00e9 s\u00edt\u011b, n\u011bkter\u00e9 z jej\u00edch mo\u017en\u00fdch akc\u00ed a sou\u0159adnice, stejn\u011b jako p\u0159edpokl\u00e1dan\u00fd v\u00fdsledek z\u00e1pasu.<\/i><\/p>\n

Toto je zvl\u00e1\u0161\u0165 relevantn\u00ed aspekt v p\u0159\u00edpad\u011b her. Obvykle se na hern\u00ed um\u011blou inteligenci nevztahuj\u00ed b\u011b\u017en\u011b pou\u017e\u00edvan\u00e9 metody strojov\u00e9ho u\u010den\u00ed. Nap\u0159\u00edklad obvykle nemus\u00ed vyhr\u00e1vat nebo d\u011blat v\u0161e, co je v jeho sil\u00e1ch pro v\u00edt\u011bzstv\u00ed, jako tomu bylo v p\u0159\u00edpad\u011b her Atari. \u010cast\u011bji je \u00faloha AI, aby hru u\u010dinila atrak\u010dn\u011bj\u0161\u00ed. M\u016f\u017ee b\u00fdt nezbytn\u00e9, aby hr\u00e1la roli a chovala se tak, jak to vy\u017eaduje charakter postavy, za kterou je zodpov\u011bdn\u00e1. Tak\u017ee hern\u00ed AI je v\u00edce spojena s hern\u00edm designem a vypr\u00e1v\u011bn\u00edm p\u0159\u00edb\u011bh\u016f a mus\u00ed m\u00edt pot\u0159ebn\u00e9 n\u00e1stroje, aby mohla \u0159\u00eddit sv\u00e9 chov\u00e1n\u00ed k dosa\u017een\u00ed stanoven\u00e9ho c\u00edle. \u010cist\u00e9 strojov\u00e9 u\u010den\u00ed pro to nen\u00ed v\u017edy vhodn\u00e9 \u2013 co\u017e znamen\u00e1, \u017ee je t\u0159eba hledat n\u011bco jin\u00e9ho. <\/p>\n

<\/p>\n

Praktick\u00e9 probl\u00e9my strojov\u00e9ho u\u010den\u00ed<\/h3>\n

\nTyto probl\u00e9my se objevily p\u0159i v\u00fdvoji um\u011bl\u00e9 inteligence zalo\u017een\u00e9 na strojov\u00e9m u\u010den\u00ed pro Unleash<\/a><\/noindex>, kde se AI mus\u00ed chovat jako b\u011b\u017en\u00ed hr\u00e1\u010di \u2013 tedy b\u00fdt stejn\u011b flexibiln\u00ed a vynal\u00e9zav\u00ed. <\/p>\n

Stejn\u011b jako Starcraft II, je Unleashed mnohem slo\u017eit\u011bj\u0161\u00ed ne\u017e \u0161achy a go na Atari. Hratelnost je intuitivn\u011b srozumiteln\u00e1 a snadno se ovl\u00e1d\u00e1, ale k opravdov\u00e9mu \u00fasp\u011bchu v n\u00ed jsou pot\u0159ebn\u00e9 ur\u010dit\u00e9 dovednosti v \u0159\u00edzen\u00ed meta. Hr\u00e1\u010d mus\u00ed vybudovat labyrinty, pos\u00edlat na nep\u0159\u00e1tele monstra a prom\u00fd\u0161let svou strategii v ekonomice, \u00fatoku a obran\u011b budov po celou dobu hry. K tomu mus\u00ed bluffovat a p\u0159edv\u00eddat tahy druh\u00fdch, stejn\u011b jako \u0159\u00eddit psychologickou metu \u2013 ta d\u011bl\u00e1 z pokeru n\u011bco v\u00edc ne\u017e jen statistickou hru.<\/p>\n

\"Jak
\nScreenshot z Unleashed<\/i><\/p>\n

P\u0159i hled\u00e1n\u00ed nejvhodn\u011bj\u0161\u00ed architektury pro tyto \u00fa\u010dely byly nejprve do hry nainstalov\u00e1ny technologie jako neuroevoluce<\/a><\/noindex> a hlubok\u00e9 u\u010den\u00ed a otestov\u00e1ny, jak se prok\u00e1\u017e\u00ed v surov\u00e9 podob\u011b jako protivn\u00edk AI. <\/p>\n

Bylo to ot\u0159esn\u00e9. <\/p>\n

Rychle se uk\u00e1zalo, \u017ee v Unleashed je nutn\u00e9 vy\u0159e\u0161it mnoho glob\u00e1ln\u00edch probl\u00e9m\u016f, na kter\u00e9 se t\u011b\u017eko zam\u011b\u0159uje strojov\u00e9 u\u010den\u00ed. <\/p>\n

Jedn\u00edm z c\u00edl\u016f je vybudov\u00e1n\u00ed efektivn\u00edho labyrintu. Stejn\u011b jako v mnoha hr\u00e1ch, jejich\u017e c\u00edlem je ochrana v\u011b\u017ee, i zde se hr\u00e1\u010di mus\u00ed kolem n\u00ed postavit labyrint, kter\u00fdm se budou prob\u00edjet p\u0159\u00ed\u0161ery. Ty je t\u0159eba odstranit pomoc\u00ed zbran\u00ed rozm\u00edst\u011bn\u00fdch v labyrintu. Ide\u00e1ln\u011b by m\u011bl b\u00fdt labyrint co nejdel\u0161\u00ed, aby bylo mo\u017en\u00e9 p\u0159\u00ed\u0161er\u00e1m zp\u016fsobit dostate\u010dn\u00e9 po\u0161kozen\u00ed a nedovolit jim dos\u00e1hnout v\u011b\u017ee. N\u011bkter\u00e9 typy zbran\u00ed jsou na p\u0159\u00ed\u0161ery \u00fa\u010dinn\u011bj\u0161\u00ed ne\u017e jin\u00e9, proto by m\u011bly b\u00fdt um\u00edst\u011bny v labyrintu co nejd\u0159\u00edve, ne\u017e ty ostatn\u00ed. Zaj\u00edmavost\u00ed Unleash je, \u017ee neexistuje dokonal\u00fd labyrint: ve h\u0159e je tolik typ\u016f p\u0159\u00ed\u0161er, \u017ee nakonec n\u011bkdo z nich bez p\u0159ek\u00e1\u017eek projde n\u011bjakou \u010d\u00e1st\u00ed labyrintu. Ka\u017ed\u00fd labyrint je t\u0159eba p\u0159izp\u016fsobit nov\u00fdm p\u0159\u00ed\u0161er\u00e1m, kter\u00e9 spou\u0161t\u011bj\u00ed jin\u00ed hr\u00e1\u010di. T\u00edm p\u00e1dem bylo nutn\u00e9 nau\u010dit um\u011blou inteligenci nejen stav\u011bt labyrinty, ale tak\u00e9 vytv\u00e1\u0159et efektivn\u00ed labyrinty pro r\u016fzn\u00e9 sc\u00e9n\u00e1\u0159e, se kter\u00fdmi se lze setkat jak v ran\u00e9, tak v pozdn\u00ed verzi hry. <\/p>\n

Um\u011bl\u00e1 inteligence se tak\u00e9 musela nau\u010dit vypo\u010d\u00edt\u00e1vat, jak\u00e9 p\u0159\u00ed\u0161ery se objev\u00ed v labyrintu. To je jak\u00fdsi inverzn\u00ed probl\u00e9m ve srovn\u00e1n\u00ed se stavbou labyrintu. Stejn\u011b jako v mnoha jin\u00fdch hr\u00e1ch, v Unleash nesta\u010d\u00ed jen zv\u00fd\u0161it arm\u00e1du a poslat ji do t\u00e1bora nep\u0159\u00edtele: je tak\u00e9 t\u0159eba \u0161pehovat nep\u0159\u00e1telskou obranu a strukturovat sv\u00e9 vojsko tak, aby co nejefektivn\u011bji ude\u0159ilo na slab\u00e1 m\u00edsta protivn\u00edka. Arm\u00e1da p\u0159\u00ed\u0161er by m\u011bla tak vz\u00e1jemn\u011b spolupracovat, aby co nej\u00fasp\u011b\u0161n\u011bji prorazila labyrint. N\u011bkdy je tak\u00e9 nutn\u00e9 vypustit p\u0159\u00ed\u0161ery v ur\u010dit\u00e9m po\u0159ad\u00ed v z\u00e1vislosti na jejich funkc\u00edch a rol\u00edch. To tak\u00e9 zvy\u0161uje mno\u017estv\u00ed r\u016fzn\u00fdch kombinac\u00ed.<\/p>\n

Kone\u010dn\u011b, proto\u017ee hr\u00e1\u010d mus\u00ed jak vytv\u00e1\u0159et labyrinty, tak shroma\u017e\u010fovat arm\u00e1du monster, mus\u00ed se AI tak\u00e9 nau\u010dit naj\u00edt rovnov\u00e1hu mezi \u00fatokem a obranou. Tak\u00e9 je d\u016fle\u017eit\u00e9 vz\u00edt v \u00favahu, \u017ee \u010d\u00edm v\u00edce hr\u00e1\u010d zvy\u0161uje arm\u00e1du monster a \u010d\u00edm v\u00edce buduje labyrint, t\u00edm v\u00edce prost\u0159edk\u016f pro to pot\u0159ebuje. Spr\u00e1vn\u00e1 \u00fato\u010dn\u00e1 strategie je proto nesm\u00edrn\u011b d\u016fle\u017eit\u00e1 jak pro ekonomiku b\u011bhem hry, tak pro v\u00edt\u011bzstv\u00ed v n\u00ed. Aby byla zaji\u0161t\u011bna sout\u011b\u017eivost, mus\u00ed b\u00fdt AI schopno um\u00eds\u0165ovat prost\u0159edky tak, aby vytvo\u0159ila silnou arm\u00e1du monster, ani\u017e by ohrozila s\u00edlu labyrintu. Investice do monster m\u016f\u017ee b\u00fdt ekonomicky v\u00fdhodn\u00e1, ale zvy\u0161uje riziko, \u017ee labyrint obsad\u00ed monstra protivn\u00edka. Pokud se naopak zam\u011b\u0159\u00edte na pos\u00edlen\u00ed obrany labyrintu, m\u016f\u017ee to oslabit va\u0161i ekonomiku. \u017d\u00e1dn\u00fd z t\u011bchto sc\u00e9n\u00e1\u0159\u016f nepovede k v\u00edt\u011bzstv\u00ed. Probl\u00e9m optimalizace v Unleashed se tedy ukazuje jako rozs\u00e1hlej\u0161\u00ed ne\u017e v p\u0159\u00edpad\u011b \u0161achu nebo Starcraftu a zahrnuje nutnost ob\u011btovat n\u011bco a posuzovat sv\u016fj zisk o n\u011bkolik krok\u016f dop\u0159edu. <\/p>\n

Jak se um\u011bl\u00e1 inteligence u\u010d\u00ed, vyvst\u00e1v\u00e1 mnoho d\u0159\u00edve nezohledn\u011bn\u00fdch probl\u00e9m\u016f. Zpo\u010d\u00e1tku AI \u010dasto dosahovala ur\u010dit\u00e9 \u00farovn\u011b v\u00fdvoje, p\u0159i kter\u00e9 za\u010d\u00ednala ch\u00e1pat n\u011bkter\u00e9 aspekty hry \u2014 nap\u0159\u00edklad, jak\u00e1 zbra\u0148 v labyrintu je \u00fa\u010dinn\u00e1 proti konkr\u00e9tn\u00edm typ\u016fm monster nebo kter\u00e9 monstra nejl\u00e9pe proch\u00e1zej\u00ed ur\u010dit\u00fdmi \u00faseky labyrintu. Ale u\u010den\u00ed bylo pomal\u00e9 a vedlo k vytv\u00e1\u0159en\u00ed jednotn\u00fdch strategi\u00ed. <\/p>\n

<\/p>\n

Nutnost paraleln\u00edch p\u0159\u00edstup\u016f<\/h3>\n

\nZat\u00edmco se v\u00fdvoj AI zalo\u017een\u00e9 na strojov\u00e9m u\u010den\u00ed posouv\u00e1 pomalu a ne zrovna \u00fasp\u011b\u0161n\u011b, pro ostatn\u00ed etapy testov\u00e1n\u00ed a v\u00fdvoje se stal pot\u0159ebn\u00fdm lep\u0161\u00ed AI a spolehliv\u011bj\u0161\u00ed konkurence AI. Pro jejich realizaci byla pou\u017eita architektura Utility, kter\u00e1 umo\u017e\u0148uje vytv\u00e1\u0159et specializovan\u00e9 AI pro testov\u00e1n\u00ed a kontrolu kvality hry, vnit\u0159n\u00ed testy a vyv\u00e1\u017een\u00ed zbran\u00ed a monster, a pro vytv\u00e1\u0159en\u00ed konkr\u00e9tn\u00edch bludi\u0161\u0165 a monster. B\u011bhem v\u00fdvoje Unleash tv\u016frci sami zdokonalovali dovednosti v jej\u00edm p\u0159ekon\u00e1n\u00ed, a pot\u00e9 se rozhodli vyu\u017e\u00edt nabyt\u00e9 znalosti pro vytvo\u0159en\u00ed slo\u017eit\u011bj\u0161\u00edho Utility AI. Tak se uk\u00e1zalo, \u017ee mnoh\u00e9 probl\u00e9my, kter\u00e9 se objevuj\u00ed v syst\u00e9mech um\u011bl\u00e9 inteligence zalo\u017een\u00e9 na strojov\u00e9m u\u010den\u00ed, lze snadno vy\u0159e\u0161it pomoc\u00ed Utility syst\u00e9m\u016f, kter\u00e9 vyu\u017e\u00edvaj\u00ed v nich zakotven\u00e9 znalosti, a naopak. <\/p>\n

Nap\u0159\u00edklad efektivn\u011bj\u0161\u00ed bludi\u0161t\u011b lze l\u00e9pe stav\u011bt pomoc\u00ed Utility AI, kter\u00e1 vych\u00e1z\u00ed z datab\u00e1z\u00ed znalost\u00ed sestaven\u00fdch na z\u00e1klad\u011b v\u00fdsledk\u016f intern\u00edch test\u016f. Algoritmus pro stavbu bludi\u0161t\u011b a um\u00edst\u011bn\u00ed zbran\u00ed v n\u011bm lze snadno popsat a naprogramovat tak, jak by to usnadnilo obran\u011b v\u011b\u017ee p\u0159ed konkr\u00e9tn\u00edmi monstry \u017eiv\u00e9mu hr\u00e1\u010di. Vytvo\u0159en\u00ed arm\u00e1dy monster na z\u00e1klad\u011b znalost\u00ed o soupe\u0159ov\u011b z\u00e1kladn\u011b se v\u0161ak uk\u00e1zalo jako slo\u017eit\u00fd \u00fakol pro takov\u00e9 AI, proto\u017ee mno\u017estv\u00ed r\u016fzn\u00fdch podm\u00ednek a kombinac\u00ed, kter\u00e9 je t\u0159eba zohlednit, ohromovalo. P\u0159i takov\u00e9m druhu architektury AI by vyhled\u00e1v\u00e1n\u00ed vhodn\u00fdch sad monster zabralo nekone\u010dn\u011b dlouh\u00fd \u010das. Pak by se s ohledem na dan\u00e1 omezen\u00ed hlubok\u00e9 u\u010den\u00ed stalo ide\u00e1ln\u00edm \u0159e\u0161en\u00edm pro tento \u00fakol. <\/p>\n

<\/p>\n

Vytv\u00e1\u0159en\u00ed hybridn\u00edch AI<\/h3>\n

\nBylo tedy rozhodnuto kombinovat dva p\u0159\u00edstupy a t\u00edm vytvo\u0159it hybridn\u00ed syst\u00e9m um\u011bl\u00e9 inteligence zalo\u017een\u00fd na strojov\u00e9m u\u010den\u00ed a Utility. My\u0161lenka spo\u010d\u00edvala v tom, \u017ee tam, kde bylo t\u0159eba zpracovat obrovsk\u00e9 mno\u017estv\u00ed kombinac\u00ed a stav\u016f hry, nebo kde bylo pot\u0159eba ji n\u011b\u010demu nau\u010dit, se vyu\u017e\u00edvalo strojov\u00e9 u\u010den\u00ed. Pro dal\u0161\u00ed \u00fakoly, kde je lep\u0161\u00ed se spolehnout na osobn\u00ed zku\u0161enosti v\u00fdvoj\u00e1\u0159\u016f, byly zapojeny syst\u00e9my Utility. V\u00fdhodou tohoto p\u0159\u00edstupu je tak\u00e9 to, \u017ee je mo\u017en\u00e9 l\u00e9pe kontrolovat chov\u00e1n\u00ed AI, aby p\u0159esn\u011bji sledovala stanoven\u00fd c\u00edl. Nap\u0159\u00edklad je mo\u017en\u00e9 vyu\u017e\u00edt Utility AI k zaji\u0161t\u011bn\u00ed rovnov\u00e1hy mezi \u00fatokem a obranou a t\u00edm vytvo\u0159it r\u016fzn\u00e9 \u00farovn\u011b agrese, nebo vytvo\u0159it r\u016fzn\u00e9 konfigurace bludi\u0161\u0165 pro r\u016fzn\u00e9 AI, aby si vytvo\u0159ily vlastn\u00ed hern\u00ed styly. Tak\u00e9 je mo\u017en\u00e9 nastavit neur\u00e1ln\u00edm s\u00edt\u00edm ur\u010dit\u00e9 hodnotov\u00e9 syst\u00e9my pro vytv\u00e1\u0159en\u00ed r\u016fzn\u00fdch preferenc\u00ed p\u0159i v\u00fdb\u011bru vzdu\u0161n\u00fdch nebo pozemn\u00edch monster, \u010d\u00edm\u017e lze tak\u00e9 jednotliv\u00fdm AI dodat osobitost. Existuje tak\u00e9 mnoho dal\u0161\u00edch mo\u017enost\u00ed pro realizaci projektov\u00fdch \u0159e\u0161en\u00ed a v\u0161echny zd\u016fraz\u0148uj\u00ed siln\u00e9 str\u00e1nky dan\u00e9ho typu architektury AI. <\/p>\n

Hybridn\u00ed p\u0159\u00edstup tak\u00e9 p\u0159edstavoval odpov\u011b\u010f na jinou ot\u00e1zku, kter\u00e1 se objevila p\u0159ed t\u00fdmem v\u00fdvoj\u00e1\u0159\u016f p\u0159i v\u00fdvoji um\u011bl\u00e9 inteligence pro Unleash: zda pou\u017e\u00edvat jednotnou glob\u00e1ln\u00ed hlubokou neur\u00e1ln\u00ed s\u00ed\u0165 zalo\u017eenou na strojov\u00e9m u\u010den\u00ed, aby zohlednila v\u0161echen vstup a v\u00fdstup, nebo je lep\u0161\u00ed navrhnout AI s hierarchickou strukturou? <\/p>\n

\"Jak
\nDva architektonick\u00e9 styly pou\u017eit\u00e9 v Unleash: vlevo - velk\u00e1 hlubok\u00e1 neur\u00e1ln\u00ed s\u00ed\u0165 s vlastn\u00ed jednotnou architekturou, vpravo - hierarchick\u00fd syst\u00e9m, v n\u011bm\u017e m\u00e1 ka\u017ed\u00e1 s\u00ed\u0165 sv\u016fj vlastn\u00ed \u00fakol.<\/i><\/p>\n

P\u0159esto by bylo dobr\u00e9 vytvo\u0159it jednotn\u00fd p\u0159\u00edstup k syst\u00e9mu um\u011bl\u00e9 inteligence, do jeho\u017e architektury v\u00fdvoj\u00e1\u0159i nezapracov\u00e1vali svou vlastn\u00ed zku\u0161enost. \u010c\u00edm v\u00edce byl hern\u00ed syst\u00e9m pou\u017e\u00edv\u00e1n, t\u00edm v\u00edce se neuronov\u00e1 s\u00ed\u0165 rozr\u016fstala. P\u0159itom nebylo mo\u017en\u00e9 odd\u011blit v\u00fduku AI a nau\u010dit je n\u011b\u010demu jednomu: bu\u010f obran\u011b, nebo \u00fatoku. A objevily se obavy, \u017ee obecn\u011bj\u0161\u00ed p\u0159\u00edstup by vedl k v\u00fdrazn\u00e9mu zv\u00fd\u0161en\u00ed v\u00fdpo\u010detn\u00edho zat\u00ed\u017een\u00ed. <\/p>\n

Odtud vznikla my\u0161lenka vytvo\u0159it hierarchickou architekturu, ve kter\u00e9 by ka\u017edou konkr\u00e9tn\u00ed \u00falohu plnila specializovan\u00e1 neuronov\u00e1 s\u00ed\u0165. Podle t\u00e9to my\u0161lenky mus\u00ed um\u011bl\u00e1 inteligence nejprve rozhodnout o rozlo\u017een\u00ed zdroj\u016f na \u00fatok (nav\u00fd\u0161en\u00ed arm\u00e1dy monster) a obranu (stavbu bludi\u0161t\u011b). Jakmile se tak rozhodne, p\u0159ech\u00e1z\u00ed na dal\u0161\u00ed \u00farove\u0148 podle sv\u00e9ho v\u00fdb\u011bru a z\u00edsk\u00e1v\u00e1 p\u0159\u00edstup k pot\u0159ebn\u00e9 \u010d\u00e1sti stavu hry, po \u010dem\u017e p\u0159ij\u00edm\u00e1 podrobn\u00e1 rozhodnut\u00ed o tom, jak\u00e9 vybrat monstra a jak\u00e9 zbran\u011b um\u00edstit do bludi\u0161t\u011b.<\/p>\n

<\/p>\n

Z\u00e1v\u011br a n\u00e1sleduj\u00edc\u00ed kroky<\/h3>\n

\nP\u0159i hybridn\u00edm p\u0159\u00edstupu Utility AI, do kter\u00e9 jsou zahrnuty s\u00edt\u011b na z\u00e1klad\u011b strojov\u00e9ho u\u010den\u00ed, p\u0159ipom\u00edn\u00e1 hierarchickou architekturu. Ta se zase podob\u00e1 biologick\u00e9mu mozku, kde r\u016fzn\u00e9 nervov\u00e9 centrum odpov\u00eddaj\u00ed ka\u017ed\u00e9 za svou \u00falohu.<\/p>\n

Sou\u010dasn\u00fd nep\u0159\u00edtel v Unleash je velmi t\u011b\u017eko poraziteln\u00fd: dok\u00e1\u017ee se p\u0159izp\u016fsobit jak\u00e9koli situaci ve h\u0159e, \u2014 p\u0159i\u010dem\u017e v\u00fdvoj\u00e1\u0159i mohou m\u011bnit jejich nastaven\u00ed podle sv\u00e9ho uv\u00e1\u017een\u00ed. Podle n\u00e1zoru autora \u010dl\u00e1nku by v pr\u016fb\u011bhu \u010dasu m\u011bl hybridn\u00ed p\u0159\u00edstup z\u00edskat v\u011bt\u0161\u00ed roz\u0161\u00ed\u0159en\u00ed a objevit se v mnoha dal\u0161\u00edch hr\u00e1ch. Mo\u017en\u00e1 se jednou poda\u0159\u00ed implementovat strojov\u011b u\u010denou um\u011blou inteligenci \u010dist\u011b do hern\u00edho procesu. Ale je jasn\u00e9, \u017ee na to je\u0161t\u011b bude zapot\u0159eb\u00ed \u010dasu. Prozat\u00edm je c\u00edlem naj\u00edt takovou architekturu, kter\u00e1 se sama p\u0159izp\u016fsob\u00ed p\u0159edlo\u017een\u00fdm \u00fakol\u016fm a nalezne optim\u00e1ln\u00ed zp\u016fsoby jejich \u0159e\u0161en\u00ed.
\n
Zdroj: habr.com<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"

\u0412 \u043f\u0440\u043e\u0434\u043e\u043b\u0436\u0435\u043d\u0438\u0435 \u043d\u0435\u043a\u043e\u0433\u0434\u0430 \u043f\u043e\u0434\u043d\u044f\u0442\u043e\u0439 \u0432 \u043d\u0430\u0448\u0435\u043c \u0431\u043b\u043e\u0433\u0435 \u0442\u0435\u043c\u044b \u0438\u0433\u0440\u043e\u0432\u043e\u0433\u043e \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0433\u043e \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442\u0430 \u043f\u043e\u0433\u043e\u0432\u043e\u0440\u0438\u043c \u043e \u0442\u043e\u043c, \u043d\u0430\u0441\u043a\u043e\u043b\u044c\u043a\u043e \u043f\u0440\u0438\u043c\u0435\u043d\u0438\u043c\u043e \u043a \u043d\u0435\u043c\u0443 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u0438 \u0432 \u043a\u0430\u043a\u043e\u043c \u0432\u0438\u0434\u0435. \u0421\u0432\u043e\u0438\u043c \u043e\u043f\u044b\u0442\u043e\u043c \u0438 \u0432\u044b\u0431\u0440\u0430\u043d\u043d\u044b\u043c\u0438 \u043d\u0430 \u0435\u0433\u043e \u043e\u0441\u043d\u043e\u0432\u0435 \u0440\u0435\u0448\u0435\u043d\u0438\u044f\u043c\u0438 \u043f\u043e\u0434\u0435\u043b\u0438\u043b\u0441\u044f \u044d\u043a\u0441\u043f\u0435\u0440\u0442 \u043f\u043e \u0432\u043e\u043f\u0440\u043e\u0441\u0430\u043c \u0418\u0418 \u0432 Apex Game Tools \u042f\u043a\u043e\u0431 \u0420\u0430\u0441\u043c\u0443\u0441\u0441\u0435\u043d. \u0412 \u043f\u043e\u0441\u043b\u0435\u0434\u043d\u0438\u0435 \u0433\u043e\u0434\u044b \u0432\u0435\u0434\u0451\u0442\u0441\u044f \u043c\u043d\u043e\u0433\u043e \u0440\u0430\u0437\u0433\u043e\u0432\u043e\u0440\u043e\u0432 \u043e \u0442\u043e\u043c, \u0447\u0442\u043e \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u043a\u0430\u0440\u0434\u0438\u043d\u0430\u043b\u044c\u043d\u043e […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[702],"tags":[],"class_list":["post-52119","post","type-post","status-publish","format-standard","hentry","category-news"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\n","aioseo_head_json":{"title":"\ud83e\udd47Jak funguje hybridn\u00ed hern\u00ed AI a v \u010dem spo\u010d\u00edvaj\u00ed jej\u00ed v\u00fdhody | ProHoster","description":"V n\u00e1vaznosti na d\u0159\u00edve prob\u00edran\u00e9 t\u00e9ma hern\u00ed um\u011bl\u00e9 inteligence v na\u0161em blogu se pod\u00edv\u00e1me na jej\u00ed aplikovatelnost.","canonical_url":"https:\/\/prohoster.info\/cs\/blog\/news\/kak-ustroen-gibridnyj-igrovoj-ii-i-v-chyom-ego-preimushhestva","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"cs_CZ","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47\u041a\u0430\u043a \u0443\u0441\u0442\u0440\u043e\u0435\u043d \u0433\u0438\u0431\u0440\u0438\u0434\u043d\u044b\u0439 \u0438\u0433\u0440\u043e\u0432\u043e\u0439 \u0418\u0418 \u0438 \u0432 \u0447\u0451\u043c \u0435\u0433\u043e \u043f\u0440\u0435\u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0430 | ProHoster","og:description":"\u0412 \u043f\u0440\u043e\u0434\u043e\u043b\u0436\u0435\u043d\u0438\u0435 \u043d\u0435\u043a\u043e\u0433\u0434\u0430 \u043f\u043e\u0434\u043d\u044f\u0442\u043e\u0439 \u0432 \u043d\u0430\u0448\u0435\u043c \u0431\u043b\u043e\u0433\u0435 \u0442\u0435\u043c\u044b \u0438\u0433\u0440\u043e\u0432\u043e\u0433\u043e \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0433\u043e \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442\u0430 \u043f\u043e\u0433\u043e\u0432\u043e\u0440\u0438\u043c \u043e \u0442\u043e\u043c, \u043d\u0430\u0441\u043a\u043e\u043b\u044c\u043a\u043e \u043f\u0440\u0438\u043c\u0435\u043d\u0438\u043c\u043e.","og:url":"https:\/\/prohoster.info\/cs\/blog\/news\/kak-ustroen-gibridnyj-igrovoj-ii-i-v-chyom-ego-preimushhestva","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2019-10-31T21:00:00+00:00","article:modified_time":"2020-02-18T10:59:47+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"52119","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":null,"schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":"2026-01-24 02:32:21","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-02-28 20:49:49","updated":"2026-01-24 02:32:21","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts\/52119","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/comments?post=52119"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts\/52119\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/media?parent=52119"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/categories?post=52119"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/tags?post=52119"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}