news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

on august 29, in a newly published paper, google researchers created a neural network that can generate real-time game graphics for the classic shooting game "doom" without using a traditional game engine.

the researchers simultaneously published this research on github and released the real-time game engine gamengen. it is reported that gamengen is the first game engine driven entirely by neural models, capable of simulating "doom" in real time at more than 20 frames per second on a single tpu (google's custom ai accelerator chip), and the generated picture quality is very close to the real game, without the need for conventional components of the game engine. the working principle of this engine is to use the diffusion model to predict each frame in real time, so the game screen is generated in real time based on the player's operation and interaction with the complex environment. gamengen's training is divided into two stages. first, the rl agent learns to play the game and records the training process, and secondly, the diffusion model is trained to generate the next frame based on the order of past frames and actions. this may mean that artificial intelligence will be able to create games on the fly and provide personalized services for each player.