Groundbreaking Release: Google Researchers Unveil the First AI Game Engine

Google Develops Neural Network for Real-Time Generation of DOOM Gameplay

On August 29, Google researchers announced a groundbreaking paper detailing a novel neural network capable of generating gameplay visuals for the classic shooter game DOOM in real time, without relying on traditional game engines.

The findings, including the real-time game engine, GameNGen, have been published on GitHub. GameNGen is hailed as the first completely neural model-driven game engine, capable of simulating DOOM at over 20 frames per second on a single TPU (Google's custom AI accelerator chip). The visuals generated closely resemble those of the actual game, all without the conventional components of a traditional game engine.

This innovative engine employs diffusion models to predict each frame in real time, dynamically adjusting the visuals based on player actions and complex environments. The training process for GameNGen consists of two phases: first, a reinforcement learning (RL) agent plays the game and collects data; second, the diffusion model is trained to generate the next frame based on the sequence of previous frames and actions. As a result, the AI can instantly create gameplay tailored to each player, offering a unique experience.

This technological advancement opens up new possibilities in game development, paving the way for increasingly personalized and interactive gaming experiences.

Most people like

Find AI tools in YBX