A Bayesian Model for Plan Recognition in RTS Games applied to StarCraft

TL;DR

This paper introduces a Bayesian model that predicts RTS game build trees from noisy data, learning from replays to enhance adaptive AI in StarCraft with minimal developer effort.

Contribution

A simple, unsupervised Bayesian approach for RTS build tree prediction that leverages player replays and requires minimal additional work from developers.

Findings

High-quality, robust predictions in StarCraft

Effective use of noisy observational data

Minimal developer effort required

Abstract

The task of keyhole (unobtrusive) plan recognition is central to adaptive game AI. "Tech trees" or "build trees" are the core of real-time strategy (RTS) game strategic (long term) planning. This paper presents a generic and simple Bayesian model for RTS build tree prediction from noisy observations, which parameters are learned from replays (game logs). This unsupervised machine learning approach involves minimal work for the game developers as it leverage players' data (com- mon in RTS). We applied it to StarCraft1 and showed that it yields high quality and robust predictions, that can feed an adaptive AI.