Google Unveils Revolutionary AI Game Engine Capable of Real-Time Generation
In a groundbreaking development, Google researchers have announced the creation of a new artificial intelligence (AI) game engine, dubbed GameNGen. This cutting-edge technology is entirely powered by a neural model, capable of real-time generation over a long trajectory, and can generate complex environments at a high number of frames.
The researchers claim that GameNGen was able to interactively simulate the classic 1993 video game Doom at more than 20 frames per second, a remarkable achievement that showcases the engine’s capabilities. This interactive simulation means that players can not only watch the generated scenes but also interact with them.
According to the researchers, two processes were followed to train the AI-powered game engine. The first involved training using Stable Diffusion v1.4, while the second employed a novel method to mitigate auto-regression drift, where Gaussian noise was added to encode the frames.
The team also used automatic reinforcement learning (RL) agents to collect data, which would have been impossible to achieve with human players. This allowed them to collect a large sample of data, further enhancing the engine’s capabilities.
Although the AI game engine is not currently available for download or testing, the research paper has been published in the online pre-print journal arXiv, along with a GitHub listing. However, it’s important to note that publishing a paper on arXiv does not require peer review, so a full evaluation of the claims and methodology is yet to be done.
The creation of a game engine is a complex task, requiring the system to generate complex 2D and 3D environments at a high speed consistently, while also ensuring logical sequencing and level progression. GameNGen’s achievement in simulating Doom at over 20 frames per second is a significant step forward in the development of AI-powered game engines.
