Mängu tegevuse lindistamisel põhinev tehisintellekti süsteem mitte-mängija tegelastele
Date
2018
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Hästi optimeeritud mitte-mängija tegelased (MMT) on vastaste või meeskonna kaaslastena üheks peamiseks osaks mitme mängija mängudes. Enamus mänguroboteid on ehitatud jäikade süsteemide peal, mis võimaldavad vaid loetud arvu otsuseid ja animatsioone. Kogenud mängijad suudavad eristada mänguroboteid inimmängijatest ning ette ennustada nende liigutusi ja strateegiaid. See alandab mängukogemuse kvaliteeti. Seetõttu, eelistavad mitme mängijaga mängude mängijad mängida pigem inimmängijate kui MMTde vastu. Virtuaalreaalsuse (VR) mängud ja VR mängijad on siiani veel väike osa mängutööstusest ja mitme mängija VR mängud kannatavad mängijabaasi kaotusest, kui mänguomanikud ei suuda leida teisi mängijaid, kellega mängida. See uurimus demonstreerib mängulindistustel põhineva tehisintellekt (TI) süsteemi rakendatavust VR esimese isiku vaates tulistamismängule Vrena. Teemamäng kasutab ebatavalist liikumisesüsteemi, milles mängijad liiguvad otsiankrute abil. VR mängijate liigutuste imiteerimiseks loodi AI süsteem, mis kasutab mängulindistusi navigeerimisandmetena. Süsteem koosneb kolmest peamisest funktsionaalsusest. Need funktsionaalsused on mängutegevuse lindistamine, andmete töötlemine ja navigeerimine. Mängu keskkond on tükeldatud kuubikujulisteks sektoriteks, et vähendada erinevate asukohal põhinevate olekute arvu ning mängutegevus on lindistatud ajaintervallide ja tegevuste põhjal. Loodud mängulogid on segmenteeritud logilõikudeks ning logilõikude abil on loodud otsingutabel. Otsingutabelit kasutatakse MMT agentide navigeerimiseks ning MMTde otsuste langetamise mehanism jäljendab olek-tegevus-tasu kontseptsiooni. Loodud töövahendi kvaliteeti hinnati uuringu põhjal, millest saadi märkimisväärset tagasisidet süsteemi täiustamiseks.
A well optimized Non-Player Character (NPC) as an opponent or a teammate is a major part of the multiplayer games. Most of the game bots are built upon a rigid system with numbered decisions and animations. Experienced players can distinguish bots from hu-man players and they can predict bot movements and strategies. This reduces the quality of the gameplay experience. Therefore, multiplayer game players favour playing against human players rather than NPCs. VR game market and VR gamers are still a small frac-tion of the game industry and multiplayer VR games suffer from loss of their player base if the game owners cannot find other players to play with. This study demonstrates the applicability of an Artificial Intelligence (AI) system based on gameplay recordings for a Virtual Reality (VR) First-person Shooter (FPS) game called Vrena. The subject game has an uncommon way of movement, in which the players use grappling hooks to navigate. To imitate VR players’ movements and gestures an AI system is developed which uses gameplay recordings as navigation data. The system contains three major functionality. These functionalities are gameplay recording, data refinement, and navigation. The game environment is sliced into cubic sectors to reduce the number of positional states and gameplay is recorded by time intervals and actions. Produced game logs are segmented into log sections and these log sections are used for creating a look-up table. The lookup table is used for navigating the NPC agent and the decision mechanism followed a way similar to the state-action-reward concept. The success of the developed tool is tested via a survey, which provided substantial feedback for improving the system.
A well optimized Non-Player Character (NPC) as an opponent or a teammate is a major part of the multiplayer games. Most of the game bots are built upon a rigid system with numbered decisions and animations. Experienced players can distinguish bots from hu-man players and they can predict bot movements and strategies. This reduces the quality of the gameplay experience. Therefore, multiplayer game players favour playing against human players rather than NPCs. VR game market and VR gamers are still a small frac-tion of the game industry and multiplayer VR games suffer from loss of their player base if the game owners cannot find other players to play with. This study demonstrates the applicability of an Artificial Intelligence (AI) system based on gameplay recordings for a Virtual Reality (VR) First-person Shooter (FPS) game called Vrena. The subject game has an uncommon way of movement, in which the players use grappling hooks to navigate. To imitate VR players’ movements and gestures an AI system is developed which uses gameplay recordings as navigation data. The system contains three major functionality. These functionalities are gameplay recording, data refinement, and navigation. The game environment is sliced into cubic sectors to reduce the number of positional states and gameplay is recorded by time intervals and actions. Produced game logs are segmented into log sections and these log sections are used for creating a look-up table. The lookup table is used for navigating the NPC agent and the decision mechanism followed a way similar to the state-action-reward concept. The success of the developed tool is tested via a survey, which provided substantial feedback for improving the system.