Doomz.io -

Ten red triangles. No. Twelve. All converging on his position. Moving fast.

DOOMZ.IO provides a rigorous platform for testing decision-making under duress. We have shown that standard RL algorithms falter when faced with shrinking boundaries and resource scarcity, often resulting in "cowardly" AI. Future work will focus on scaling DOOMZ.IO to massive 100-agent simulations to study emergent swarm behaviors, coalition forming, and betrayal dynamics in open-source play.

DOOMZ.IO is a top-down, 2D multi-agent environment implemented in Unity and wrapped with a Gym-compatible API. The simulation is discrete but high-frequency (20 ticks per second). doomz.io

The lead Viper—a sleek black tank with a gold skull decal—paused to collect the crate.

The tower collapsed with a screech of tearing metal, sending a tidal wave of brown sludge across the road. The Vipers scattered, their formation shattered. One slid into a ditch. Another spun out, tracks whirring uselessly. The gold-skull Viper dodged the sludge but reversed straight into Leo's line of sight. Ten red triangles

DOOMZ.IO is a proposed browser-based and API-accessible benchmark that merges the high-octane mechanics of First-Person Shooters (FPS) with the resource-scarcity mechanics of "Battle Royale" and ".io" genre games. The core contribution of this paper is the definition of the DOOMZ.IO environment and the analysis of agent failure modes when subjected to the "Doom Clock"—a mechanic where the playable area shrinks as resources deplete.

A rocket to the ditch-bound Viper. Boom. All converging on his position

Players must hit rocks and trees to collect stone and wood. These materials are the foundation for everything from simple defensive walls to complex machinery.

DOOMZ.IO: A Multi-Agent Reinforcement Learning Benchmark for Strategic Survival Under Exponential Stress