A Unified Pipeline to Learn Dota 2 Draft Strategies
Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · 08.11.2026 · 22 (1) · 44-53
DOI: 10.1609/aiide.v22i1.43563 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, online erschienen am 08.11.2026. Verfasst von Dengfangzhou Cheng und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Like many multiplayer online battle arena games, Valve Software's Dota 2 begins with a crucial stage of alternating strategic bans and selections ("picks") of heroes to play in the game. We present a framework for automating these ban/pick choices which first trains an oracle to predict game outcomes given a full set of picks from both team, and then uses one of several methods to make interleaved ban/pick actions based on the oracle. These methods apply Monte Carlo Tree Search, coevolutionary optimization, and transformer neural network architectures. We test these methods against one another, and compare them, and ensembles of them, to the existing literature.
Zitieren
APA
Cheng, D., & Luke, S. (2026). A Unified Pipeline to Learn Dota 2 Draft Strategies. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 22(1), 44-53. https://doi.org/10.1609/aiide.v22i1.43563
Vancouver
Cheng D, Luke S. A Unified Pipeline to Learn Dota 2 Draft Strategies. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. 2026;22(1):44-53. doi:10.1609/aiide.v22i1.43563
BibTeX
@article{Cheng2026,
author = {Cheng, Dengfangzhou and Luke, Sean},
title = {A Unified Pipeline to Learn Dota 2 Draft Strategies},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment},
year = {2026},
volume = {22},
number = {1},
pages = {44--53},
doi = {10.1609/aiide.v22i1.43563}
}
EUVIMED