A Benchmark for Generalizing Across Diverse Team Strategies in Competitive Pokémon
Paper • 2506.10326 • Published
This repo stores .zip checkpoints and TensorBoard logs from vgc-bench.
The training methods are described in VGC-Bench: Towards Mastering Diverse Team Strategies in Competitive Pokemon.
results/ — Main benchmark results
Results from the main VGC-Bench experiments with varied team pool sizes.
saves_bc/ — Behavior cloning (BC) checkpointsseed1/ — BC policy checkpoints (epochs 0–100)logs_bc/ — BC TensorBoard logsseed1/saves_bc_sp/ — BC-initialized self-play (BCSP) checkpointsreg_g/{1,4,16,64}_teams/seed1/ — BCSP checkpoints across team pool sizeslogs_bc_sp/ — BCSP TensorBoard logsreg_g/{1,4,16,64}_teams/seed1_0/results_<tournament>_finals/ — VGC tournament matchup results
Each directory contains results from training on a specific VGC tournament finals matchup, with team1.txt and team2.txt defining the two teams.
| Directory | Method | Reg | Teams |
|---|---|---|---|
results_worlds_2024_finals/ |
BC + self-play, no mirrors | g | 2 |
results_atlanta_2025_finals/ |
BC + self-play, no mirrors | g | 2 |
results_euic_2025_finals/ |
BC + self-play, no mirrors | g | 2 |
results_san_antonio_2025_finals/ |
BC + self-play, no mirrors | g | 2 |
results_euic_2026_finals/ |
BC + self-play, no mirrors | f | 2 |
Each contains:
saves_bc_sp_xm/ — BCSP checkpoints (no mirror matches)reg_{g,f}/{N}_teams/seed1/logs_bc_sp_xm/ — TensorBoard logsreg_{g,f}/{N}_teams/seed1_0/team1.txt, team2.txt — Team strings for the matchupCheckpoints are Stable Baselines 3 .zip files loadable with:
from stable_baselines3 import PPO
model = PPO.load("path/to/checkpoint.zip")
See the vgc-bench README for full training and evaluation instructions.
@inproceedings{anglissvgc,
title={VGC-Bench: Towards Mastering Diverse Team Strategies in Competitive Pok{\'e}mon},
author={Angliss, Cameron L and Cui, Jiaxun and Hu, Jiaheng and Rahman, Arrasy and Stone, Peter},
booktitle={The 25th International Conference on Autonomous Agents and Multi-Agent Systems}
}