The Study On Multi-agent Decision Model In FPS Games

With rapid growth, the game industry has exceeded the film andentertainment industry in the annual output value. At the same time, the gameartificial intelligence has become as important as the graphics and soundeffect in the game design. Especially in FPS games, intelligent non-playercharacters NPCs (including enemy units, our units and neutral units) canobtain information, perceive changes in the environment around, and make arational decision-making and response. The player can feel collaboration orconfrontation with the intelligent non-player characters, which will greatlyimprove the game experience.In traditional FPS games there will be multiple non-player characters.They tend to act alone, lack cooperation, which is not in line with the realworld situation. In addition, under conventional keyboard or joystick-basedhuman-computer interaction, it is difficult for the player to command virtualteammates through the player’s speech and action just like in the realenvironment, and it is also difficult to control the player’s substitute in thegame scene. The thesis studied the multi-agent decision model for virtualcharacters FPS games, and a new model of multi-agent was proposed. Thenew model supports coordination among virtual characters. It supportscollaboration between the virtual characters and the player can interactivelycontrol virtual characters.The main research and contributions of the thesis are as following:(1) A state-based hierarchical decision-making model was proposed. Theagents are organized into multiple layers. Each layer can do autonomicdecision-making based on finite state machine. The information is gatheredupward, while the decision is made downward through the layers, to facilitatecentralized control. Decision of the upper layer, can be performed bychanging the state of the next lower layer. (2) A method was designed to interpret and execute semantic-basedspeech and action commands from the player. The paper analyzed thecommand processing procedure and broke it into five parts: instructionsynthesis, instructions packaging, instruction messaging, and instructiontranslation and instruction execution. The work studied the problems in everypart and gave one proper solution.(3) On the basis of the above work, a unified command-supportedmulti-agent hierarchical decision-making model was proposed. The modelmakes the interpreted players’ commands interactively flow from the team tothe individual soldier, and get executed by state change for virtual characters.(4) An approach to initialize the layered agents was designed, whichprovide easy modules reuse and agents extension. The method can configurethe control components and FSM of each AI layer through XML files. Themethod supports the expansion of new intelligent units by changing theconfiguration of the script control components in the XML configuration file.Game developers can easily provide new decision strategy for the agents bychanging the configuration of the finite state machine in the XMLconfiguration file. This method best reuse the script control components andstate modules.(5) Experiment and application testing has validated the effectivity of thecommand-supported multi-agent hierarchical decision-making model. Theresults stated that the decision-making model performed good to drivemultiple virtual characters, showed the cooperation between the agents, andthe command support was accurate. The whole AI model ran stable with lessthan16%of the CPU time consumption.