FootsiesGym: A New Benchmark for Two-Player Imperfect-Information Games

Chase McDonald, Nathan Tsang, Wesley N. Kerr· July 8, 2026 View original

Summary

FootsiesGym is an open-source environment for training AI in complex two-player, zero-sum, imperfect-information fighting games, designed for efficient analysis and high-throughput training. It isolates strategic interactions of fighting game neutral play, making it accessible for reinforcement learning research.

Researchers have introduced FootsiesGym, an open-source simulation environment tailored for advancing AI in two-player, zero-sum games with imperfect information. Built upon the minimalist 2D fighting game Footsies, this platform distills the core strategic elements of fighting game "neutral play," which involves cyclic and non-transitive interactions. The environment is designed for efficiency, featuring a vectorized simulator that supports high-throughput training on standard hardware, thus promoting accessibility and reproducibility in AI research. The creators have detailed its design, benchmarked several reinforcement learning algorithms within it, and outlined future research avenues it enables.

Why it matters

This environment offers a standardized, accessible platform for developing and testing advanced AI agents in complex strategic scenarios, which can lead to breakthroughs in game AI and general reinforcement learning.

How to implement this in your domain

  1. 1Explore the FootsiesGym GitHub repository to understand its architecture and API.
  2. 2Integrate the environment into existing reinforcement learning frameworks for agent development.
  3. 3Benchmark novel AI algorithms against established baselines provided within the gym.
  4. 4Contribute to the open-source project by developing new features or improving existing components.

Who benefits

GamingAI ResearchRobotics

Key takeaways

  • FootsiesGym provides a new open-source benchmark for imperfect-information games.
  • It simplifies complex fighting game mechanics for efficient AI training.
  • The environment supports high-throughput reinforcement learning on standard hardware.
  • It enables research into strategic interactions and generalizable AI agents.

Original post by Chase McDonald, Nathan Tsang, Wesley N. Kerr

"arXiv:2607.06514v1 Announce Type: new Abstract: We present FootsiesGym, an open-source environment for learning in a non-trivial two-player, zero-sum, imperfect-information game. Built on HiFight's minimalist 2D fighting game Footsies, it isolates the cyclic, non-transitive strat…"

View on X

Originally posted by Chase McDonald, Nathan Tsang, Wesley N. Kerr on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses