Pokémon Red via Reinforcement Learning

IEEE Conference on Games (CoG), pp. 1–8

Abstract

We present a Deep Reinforcement Learning (DRL) agent that successfully completes the first several hours of Pokémon Red, a classic Game Boy JRPG that exposes significant challenges as a testbed for agents, including multitasking, long horizons of tens of thousands of steps, hard exploration, and a vast array of potential policies. Our agent completes an initial segment of the game, up to Cerulean City, a location requiring progression through two cities, battle-filled passages, a maze-like cave, and defeating the first gym leader. Our experiments include various ablations that reveal vulnerabilities in reward shaping. We argue that long-form games like Pokémon hold strong potential for future research, presenting long-term coherent reasoning challenges absent from simpler arcade games. Our environment wrapper, training algorithm, human replay data, and pretrained agent are available at (REDACTED FOR REVIEW).

Authors 6

1. Marco Pleines

   TU Dortmund University

   Affiliation as printed

   TU Dortmund University,Dortmund,Germany

2. Daniel Addis
   Affiliation as printed

   Independent Researcher,Minneapolis,MN,USA

3. David Rubinstein
   Affiliation as printed

   Independent Researcher,Brooklyn,NY,USA

4. Fränk Zimmer

   Rhine-Waal University of Applied Sciences

   Affiliation as printed

   Rhine-Waal University of Applied Sciences,Kamp-Linfort,Germany

5. Mike Preuß Aachen

   Leiden University

   Affiliation as printed

   LIACS Universiteit Leiden,Leiden,Netherlands

6. Peter J. Whidden
   Affiliation as printed

   Independent Researcher,Brooklyn,NY,USA