Doomz.io Today

Our initial experiments revealed a critical failure mode in standard DQN and A3C agents. Due to the penalty for taking damage and the stochastic nature of combat, standard agents quickly learned that the optimal short-term strategy for survival was to retreat to map corners and minimize movement to avoid detection.

The arena grew quiet.

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

The explosion was a beautiful orange blossom in the gray world. +250 points flashed on Leo’s HUD.

But Leo was tired of running.

The last Viper panicked. It tried to flee, weaving erratically. Leo aimed not at the tank, but at the road ahead. A cannon shell shattered the asphalt, creating a ramp. The Viper hit it, went airborne, and flipped. It landed on its turret.

Three shots. The first took off its plasma gun. The second cracked its armor. The third—the third found the fuel cell. Our initial experiments revealed a critical failure mode

Now.