Skip to content

Wargames Training

Reinforcement learning research platform for Napoleonic-era battalion simulation.

PyPI License: MIT CI

  • Gymnasium Environments

    Fully-featured Gymnasium and PettingZoo environments spanning single battalions through corps-level operations. Observations normalized, angles as (cos θ, sin θ), positions normalized by map size.

    Environment Spec

  • PPO · Self-Play · League

    Complete training pipelines: PPO baseline, MAPPO multi-agent, self-play loop, and AlphaStar-style league training with PFSP matchmaking and Elo tracking.

    Training Guide

  • WargamesBench

    20 canonical scenarios for reproducible comparison. Results guaranteed within ± 2 % win rate across identical seeds. Open leaderboard included.

    Benchmark Leaderboard

  • GIS Terrain

    Real geographic battlefield data for Waterloo, Austerlitz, Borodino, and Salamanca. Elevation grids drive line-of-sight, movement cost, and terrain bonuses.

    Historical Scenarios

Quick Start

pip install wargames-training

git clone https://github.com/B9android/wargames_training.git
cd wargames_training
pip install -e ".[dev]"

from envs import BattalionEnv
from training.train import train
import wandb

wandb.init(project="wargames_training", config={"total_timesteps": 100_000})

env = BattalionEnv()
train(env=env, total_timesteps=100_000)

Or via CLI:

python -m training.train --config configs/experiment_1.yaml

wargames-bench --episodes 100 --label "my_policy_v1"

Results are written to docs/wargames_bench_leaderboard.md. Submit a PR to add your row to the public leaderboard.

Project Status

Version Theme Status
v1 Foundation — 1v1 battalion ✅ Complete
v2 Multi-Agent — MARL 2v2+ ✅ Complete
v3 Hierarchy — Brigade / Division HRL ✅ Complete
v4 League — AlphaStar-style training ✅ Complete
v5 Real-World Interface & Analysis ✅ Complete
v6 Physics-Accurate Simulation 🔲 Planned
v7 Operational Scale (Corps / Army) 🔲 Planned
v8 Transformer Policy & Architecture 🔲 Planned
v9 Human-in-the-Loop & Decision Support 🔲 Planned
v10 Multi-Domain & Joint Operations 🔲 Planned
v11 Real-World Data & Transfer 🔲 Planned
v12 Foundation Model & Open Platform 🔲 Planned

See the full Roadmap for milestone detail and epic breakdown.

Community

  • GitHub Discussions — research questions, benchmark results, ideas
  • Issues — bug reports and feature requests
  • Contributing — see CONTRIBUTING.md
  • Code of Conduct — see CODE_OF_CONDUCT.md

Citation

If you use Wargames Training or WargamesBench in your research, please cite:

@software{wargames_training,
  title   = {Wargames Training: An Open RL Research Platform for Military Simulation},
  author  = {Wargames Training Contributors},
  year    = {2026},
  url     = {https://github.com/B9android/wargames_training},
}