MappyLand: Fast, Accurate Mapping for Console Games

Authors

  • Joseph C. Osborn Pomona College
  • Adam Summerville California State Polytechnic University, Pomona
  • Nathan Dailey Pomona College
  • Soksamnang Lim Pomona College

Keywords:

Automated Game Design Learning, Knowledge Representation, Localization, Commonsense AI, Operational Logics, Design Recovery, Reverse Engineering

Abstract

We present MappyLand, a rewrite and enhancement of the earlier Mappy automatic game mapping system, which leverages instrumentation of a game console emulator to produce, from a sequence of game inputs, accurate annotations for action/adventure games including object detection and tracking, in-game camera movement, grid-based tile maps, and links between identified disparate spaces. The overhead of generating these annotations is on the order of one millisecond per observed frame. We also show a higher latency (but still online) algorithm for merging together previous observations of distinct game maps for the purposes of agent localization across a long period of time. Specifically, our system can determine a consistent graph of game rooms from a set of strings of game rooms, capturing behaviors like backtracking and synthesizing observations from multiple play sessions. Altogether, this fast, accurate approach to mapping yields new and useful knowledge representations and expedient ways to produce new datasets given just a game and some example play.

Downloads

Published

2021-10-04

How to Cite

Osborn, J. C., Summerville, A., Dailey, N., & Lim, S. (2021). MappyLand: Fast, Accurate Mapping for Console Games. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 17(1), 66-73. Retrieved from https://ojs.aaai.org/index.php/AIIDE/article/view/18892