MiniRTS: Strategy Identification using Bayesian Grid Models
Author
Dancker-Jensen, Mads
Term
4. term
Education
Publication year
2008
Pages
52
Abstract
This thesis investigates whether Bayesian Grid Models can support strategy identification in real-time strategy (RTS) games. To enable controlled study, the author implements MiniRTS, a simplified yet representative RTS environment that reproduces core mechanics such as units, pathfinding, groups, orders, and fog of war. On top of this environment, the work formalizes strategy identification and management, and develops Bayesian-network-based models—including a Naive Bayes identifier and tree-based learning and identification components—to infer player strategies from grid-based spatial observations. A suite of tools (model generator, strategy monitor, evaluator) supports model construction, training, and testing. The experiments comprise scenarios that compare model configurations (e.g., grid sizes and numbers of states) and assess robustness under partial observability (fog of war). Although the excerpt does not provide quantitative results, the report analyzes comparative model behavior, highlights factors influencing performance, and outlines directions for future improvements.
Denne rapport undersøger, om Bayesian Grid Models kan bruges til strategigenkendelse i real‑time strategy (RTS) spil. For at muliggøre kontrollerede studier er der implementeret et simpelt, men repræsentativt testmiljø, MiniRTS, der efterligner centrale mekanikker som enheder, pathfinding, grupper, ordrer og fog of war. Oven på dette miljø formaliseres strategigenkendelse og -styring, og der udvikles bayesianske modeller—herunder en naiv Bayes‑identifikator og træbaserede lærings‑ og identifikationskomponenter—til at udlede spillerstrategier ud fra rumlige observationer på et gitter. Et værktøjssæt (modelgenerator, strategi‑monitor og evaluator) understøtter opbygning, træning og test af modellerne. Eksperimenterne består af scenarier, der sammenligner modelkonfigurationer (f.eks. gitterstørrelser og antal tilstande) og undersøger robusthed under delvis observabilitet (fog of war). Uddraget indeholder ikke numeriske resultater, men rapporten analyserer de relative forskelle mellem modellerne, identificerer faktorer der påvirker ydeevnen, og skitserer potentielle forbedringer.
[This apstract has been generated with the help of AI directly from the project full text]