TEAMWORK UNDER EXTREME UNCERTAINTY: AI FOR POKÉMON RANKS 33 RD IN THE WORLD

Oct 12, 2026

Uploaded by Ivy

Pokémon is the highest-grossing media franchise of all time, and building a strong AI agent for its battles is exceptionally difficult due to random factors and imperfect information. The paper explains game mechanics, analyzes the game from an AI perspective, then introduces new algorithms focused on two core challenges: maintaining a balanced team and handling three distinct uncertainty sources. Evaluation is difficult, yet the resulting agent outperformed prior approaches and reached 33rd globally in a popular battle format using only 4 single-socket servers.

LanguageEnglish

CategoryGeneral

FormatPDF

Pages14

Reading Time5 min

AccessFree to download

Introduction

Background

Game analysis from an AI perspective

Methods of this paper

Experiments

arXiv :2212 . 13338v2 [ cs .AI] 4 Jan 2023 TEAMWORK UNDER EXTREME UNCERTAINTY: AI FOR POKÉMON RANKS 33 RD IN THE WORLD Nicholas R. Sarantinos University of Cambridge [nrs38@cantab.ac.uk](nrs38@cantab.ac.uk) ABSTRACT The highest grossing media franchise of all times, with over $90 billion in total revenue, is Pokémon. The video games belong to the class of Japanese Role Playing Games (J-RPG) . Developing a powerful AI agent for these games is very hard because they present big challenges to MinMax, Monte Carlo Tree Search and statistical Machine Learning, as they are vastly different from the well explored in AI literature games. An AI agent for one of these games means signiﬁcant progress in AI agents for the entire class. Further, the key principles of such work can hopefully inspire approaches to several domains that require excellent teamwork under conditions of extreme uncertainty, including managing a team of doctors, robots or employees in an ever changing environment, like a pandemic stricken region or a war-zone. In this paper we ﬁrst explain the mechanics of the game and we perform a game analysis. We continue by proposing unique AI algorithms based on our understanding that the two biggest challenges in the game are keeping a balanced team and dealing with three sources of uncertainty. Later on, we describe why evaluating the performance of such agents is challenging and we present the results of our approach. Our AI agent performed signiﬁcantly better than all previous attempts and peaked at the 33rd place in the world, in one of the most popular battle formats, while running on only 4 single socket servers. 1 Introduction Since the early days of Artiﬁcial Intelligence, a lot of research was focused on developing intelligent agents for games. Despite the signiﬁcant amount of work done in deterministic and perfect information games, including Chess [1] and GO [2], there is limited research in games like Pokémon, that combine a signiﬁcant random factor with imperfect information, outside of Poker. Pokémon, the highest grossing media franchise of all times, long ago inspired fans to develop intelligent agents capable of competing with human players [3][4][5][6] . The franchise began in 1995 with the introduction of the ﬁrst Pokémon games for the Nintendo Game Boy platform. In these games players explore a vast game world, collect, train, evolve and trade creatures known as Pokémon as well as battle against other players using a team of Pokémon. Since 1995 over 70 Pokémon games have been released and together with merchandise, movies, TV shows, cards and more they grossed over $90 billion in total revenue. The Pokémon games can be classiﬁed as either core series [7] or spin-offs [8] . The differences between the two categories are outside the context of this work and here we focus only on the former. On 18 November 2022 the ﬁrst 9th generation core series games, Pokémon Scarlet and Pokémon Violet, were released. When this work took place there were 8 generations of games and the latest core series games were Pokémon Sword and Pokémon Shield. Screenshots of these games are in Figure 1 . While every new generation brings changes, the games always allowed two players to connect their consoles locally and battle against each other. Since 2006, players can do the same via Internet. In 2009 and every year afterwards, the game publisher, Nintendo, organizes the “Video Game Championships (VGC)” where the top players across the globe are invited to battle against each other for the title of World Champion. This and other events led to the emergence of professional Pokémon players. Nintendo designed the games to be easy for young kids to play but difﬁcult for professional players to master as the battle mechanics are very complicated. Numerous fan made websites provide extensive analysis of the game mechanics, in order to aid professional and amateur players [9][10][11] . (a) A player is exploring the vast g

FAQs

Why is creating an AI agent for Pokémon battles challenging?

Because Pokémon battles combine significant randomness with imperfect information, which differs from well-studied games like Chess or GO and makes classic AI approaches harder to apply.

What are the paper’s two biggest challenges for the AI agent?

Keeping a balanced team and dealing with three sources of uncertainty.

How did the proposed AI agent perform in evaluation?

It performed significantly better than previous attempts and peaked at 33rd in the world in a popular battle format, running on only 4 single-socket servers.

Oct 12, 2026

Uploaded by Ivy