Strategic Cooperation in First-Person Shooters
Authors
Ringgaard, Alex Vendelbo ; Thomsen, Martin Guld
Term
4. term
Education
Publication year
2004
Abstract
Denne rapport introducerer to metoder, der hjælper computerstyrede holdkammerater (agenter) med at samarbejde uden at skulle forhandle. Det er vigtigt i hurtige førstepersonsskydespil, hvor mennesker kun har tid til at sende korte beskeder. Den første metode, forudsigelse, lader en agent afgøre, hvilke handlinger andre agenter har taget eller sandsynligvis vil tage. Fordi den ikke afhænger af kommunikation, er den den primære måde, vi understøtter samarbejde på. Den anden metode, værdien af kommunikation, vurderer om det kan betale sig at kommunikere i en given situation—både når der anmodes om information fra en anden agent, og når der svares på en anmodning. Den bruges som sekundær støtte til forudsigelse og kræver begrænset beskedudveksling. Selvom metoderne er generelle, fokuserer vi på strategisk samarbejde mellem forsvarende agenter i spillet Team Fortress. En empirisk evaluering viser, at teknikkerne forbedrer en agents evne til at nå sine mål. Vi konkluderer, at forudsigelse styrker agenternes præstationer, og at værdien af kommunikation finder de mest fordelagtige tidspunkter at anmode om information.
This report introduces two methods that help computer-controlled teammates (agents) cooperate without having to negotiate. This matters in fast first-person shooter games, where human players only have time to send short messages. The first method, prediction, lets an agent infer what actions other agents have taken or are likely to take. Because it does not depend on messaging, it is the primary way we support cooperation. The second method, value of communication, estimates whether it is worthwhile to communicate in a given situation—both when requesting information from another agent and when answering a request. It serves as a secondary support to prediction and requires limited message passing. Although the methods are general, we focus on strategic cooperation among defending agents in the game Team Fortress. An empirical evaluation shows that the techniques improve an agent’s ability to achieve its goals. We conclude that prediction strengthens agent performance, and that value of communication identifies the most profitable times to request information from others.
[This abstract was generated with the help of AI]
Documents