English

Territory Paint Wars: Diagnosing and Mitigating Failure Modes in Competitive Multi-Agent PPO

Machine Learning 2026-04-08 v1

Abstract

We present Territory Paint Wars, a minimal competitive multi-agent reinforcement learning environment implemented in Unity, and use it to systematically investigate failure modes of Proximal Policy Optimisation (PPO) under self-play. A first agent trained for 84,00084{,}000 episodes achieves only 26.8%26.8\% win rate against a uniformly-random opponent in a symmetric zero-sum game. Through controlled ablations we identify five implementation-level failure modes -- reward-scale imbalance, missing terminal signal, ineffective long-horizon credit assignment, unnormalised observations, and incorrect win detection -- each of which contributes critically to this failure in this setting. After correcting these issues, we uncover a distinct emergent pathology: competitive overfitting, where co-adapting agents maintain stable self-play performance while generalisation win rate collapses from 73.5%73.5\% to 21.6%21.6\%. Critically, this failure is undetectable via standard self-play metrics: both agents co-adapt equally, so the self-play win rate remains near 50%50\% throughout the collapse. We propose a minimal intervention -- opponent mixing, where 20%20\% of training episodes substitute a fixed uniformly-random policy for the co-adaptive opponent -- which mitigates competitive overfitting and restores generalisation to 77.1%77.1\% (±12.6%\pm 12.6\%, 1010 seeds) without population-based training or additional infrastructure. We open-source Territory Paint Wars to provide a reproducible benchmark for studying competitive MARL failure modes.

Keywords

Cite

@article{arxiv.2604.04983,
  title  = {Territory Paint Wars: Diagnosing and Mitigating Failure Modes in Competitive Multi-Agent PPO},
  author = {Diyansha Singh},
  journal= {arXiv preprint arXiv:2604.04983},
  year   = {2026}
}

Comments

16 pages, 5 figures

R2 v1 2026-07-01T11:55:47.637Z