Microsoft Releases AI-Generated Playable Demo of Quake II Game
Last week, Microsoft launched an interactive real-time gameplay experience of Quake II through its Copilot Labs. This innovative project leverages the company’s newly developed Muse AI models and introduces a novel approach known as the World and Human Action MaskGIT Model (WHAMM). Currently available as a research preview, the demo allows users to explore the game’s world generation and mechanics, although it comes with certain limitations in gameplay.
Innovative AI Technology Behind Quake II
In a recent blog post, Microsoft researchers elaborated on the AI-generated gameplay and the technology that powers it. The integration of AI in 2D and 3D game generation has become a focal point for researchers, as it demonstrates the potential of AI to create real-time environments and adapt to various mechanics typically employed by human players. This project serves as a testing ground for training AI models to perform real-world tasks, such as controlling robots in physical environments. Originally released in 1997, Quake II is a first-person shooter published by Activision, which is owned by Microsoft. The game features a 3D forward-scrolling format with a variety of mechanics, including jumping, crouching, shooting, and environmental destruction. Users can currently experience a single level of the game for approximately two minutes, utilizing either a controller or keyboard.
Advancements with WHAMM Technology
The development of this AI gameplay experience involved the use of Muse AI models alongside the World and Human Action Model (WHAM). The WHAMM approach is a significant upgrade from its predecessor, WHAM-1.6B, boasting the capability to generate over 10 frames per second, thus enabling real-time video generation. The gameplay resolution is set at 640×360 pixels, and one of the standout improvements in WHAMM’s performance is attributed to the implementation of the MaskGIT (Mask Generative Image Transformer) setup, which has drastically increased frame rates from one frame per second to over ten. The MaskGIT setup allows researchers to generate all tokens for an image with a limited number of forward passes. This advancement enables the AI model to predict potential moves of a single masked image in real time, resulting in a smoother gameplay experience. While the core mechanics remain true to the original Quake II, Microsoft has acknowledged several limitations in the current demo.
Limitations and Challenges in AI Gameplay
Despite the innovative technology, Microsoft has outlined several limitations associated with the AI-generated gameplay. Since the game environment is created using AI, it serves as an approximation rather than an exact replica of the real world. Players may encounter issues such as fuzzy image generations during enemy interactions and inaccuracies in combat scenarios. The WHAMM model currently operates with a context window of 0.9 seconds (equivalent to 9 frames at 10fps). Consequently, the model may lose track of objects that move out of view for longer than this duration. This can lead to unexpected experiences, such as players turning around to discover entirely new areas or looking up at the sky only to find themselves relocated within the game map. Additionally, the demo is experiencing notable latency due to its accessibility to the public.
Observer Voice is the one stop site for National, International news, Sports, Editor’s Choice, Art/culture contents, Quotes and much more. We also cover historical contents. Historical contents includes World History, Indian History, and what happened today. The website also covers Entertainment across the India and World.
