PokeRL: Reinforcement Learning for Pokemon Red

TL;DR

PokeRL introduces a modular reinforcement learning system for Pokemon Red, addressing challenges like partial observability and action loops, to train agents on early game tasks with improved robustness.

Contribution

The paper presents a loop-aware environment wrapper, anti-loop mechanisms, and hierarchical rewards, advancing RL training in complex, long-horizon games like Pokemon Red.

Findings

Successfully trained agents to complete early game tasks in Pokemon Red.

Implemented anti-loop and anti-spam mechanisms to improve training stability.

Provided open-source code for the PokeRL system.

Abstract

Pokemon Red is a long-horizon JRPG with sparse rewards, partial observability, and quirky control mechanics that make it a challenging benchmark for reinforcement learning. While recent work has shown that PPO agents can clear the first two gyms using heavy reward shaping and engineered observations, training remains brittle in practice, with agents often degenerating into action loops, menu spam, or unproductive wandering. In this paper, we present PokeRL, a modular system that trains deep reinforcement learning agents to complete early game tasks in Pokemon Red, including exiting the player's house, exploring Pallet Town to reach tall grass, and winning the first rival battle. Our main contributions are a loop-aware environment wrapper around the PyBoy emulator with map masking, a multi-layer anti-loop and anti-spam mechanism, and a dense hierarchical reward design. We argue that practical systems like PokeRL, which explicitly model failure modes such as loops and spam, are a necessary intermediate step between toy benchmarks and full Pokemon League champion agents. Code is available at https://github.com/reddheeraj/PokemonRL