FootsiesGym: A Fighting Game Benchmark for Two-Player Zero-Sum Imperfect-Information Games

2026-07-07 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Gaming & Interactive Media · Depth: Advanced, medium

Summary

FootsiesGym is an open-source environment designed for reinforcement learning in two-player, zero-sum, imperfect-information games. Built upon the minimalist 2D fighting game Footsies, this benchmark specifically isolates the cyclic, non-transitive strategic interactions characteristic of fighting game "neutral play," while maintaining sufficient simplicity for efficient analytical study. The environment includes a vectorized simulator, enabling high-throughput training on standard hardware, which enhances its accessibility and reproducibility for researchers. The authors, Chase McDonald, Nathan Tsang, and Wesley N. Kerr, describe the environment's design, present benchmarks for several reinforcement learning algorithms, and outline future research directions it facilitates. The code is publicly available on GitHub.

Key takeaway

For machine learning engineers developing agents for complex strategic games, FootsiesGym offers a valuable, accessible testbed. You can efficiently benchmark reinforcement learning algorithms against a simplified yet challenging two-player, zero-sum, imperfect-information environment. This allows you to rapidly iterate on agent designs and analyze performance in cyclic, non-transitive interactions without the overhead of full-scale fighting games. Consider integrating this open-source tool to accelerate your research into robust game AI.

Key insights

FootsiesGym provides a simplified, high-throughput environment for studying complex strategic interactions in imperfect-information games.

Principles

• Isolate core strategic interactions for focused analysis.
• Prioritize accessibility and reproducibility in environment design.
• Vectorized simulation boosts training throughput for RL algorithms.

Method

The environment design involves building on a minimalist 2D fighting game, isolating specific strategic interactions, and implementing a vectorized simulator for efficient RL algorithm benchmarking.

In practice

• Benchmark RL algorithms on fighting game "neutral play" scenarios.
• Develop agents for imperfect-information game settings.
• Analyze cyclic, non-transitive game dynamics efficiently.

Topics

• Reinforcement Learning
• Game AI
• Imperfect Information Games
• Zero-Sum Games
• FootsiesGym
• Benchmarking

Code references

• como-research/FootsiesGym
• ghimiremukesh/cams
• hly1998/ChronoPlay

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Bench-MFG: A Benchmark Suite for Learning in Stationary Mean Field Games — Bench-MFG standardizes Mean Field Game evaluation through a diverse benchmark suite and procedural generation.
• Valet: A Standardized Testbed of Traditional Imperfect-Information Card Games — Valet standardizes imperfect-information card games for robust AI algorithm benchmarking.
• In game theory, generalists sometimes win out over specialists — General-purpose policy gradient methods can surprisingly outperform specialized game-theoretic algorithms in imperfect-information games.
• ShopGym: An Integrated Framework for Realistic Simulation and Scalable Benchmarking of E-Commerce Web Agents — ShopGym offers a scalable framework for creating realistic, controllable, and reproducible e-commerce web agent evaluation environments.
• Mahjax: A GPU-Accelerated Mahjong Simulator for Reinforcement Learning in JAX — Mahjax provides a high-throughput, GPU-accelerated JAX environment for *tabula rasa* reinforcement learning in complex imperfect-information games.
• Age of LLM: A Strategic 1v1 Benchmark for Reasoning, Diplomacy and Reliability of Large Language Models under Fog of War — Benchmarking LLMs in a strategic game reveals their reasoning, diplomatic, and reliability limitations under adversarial conditions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.