The editor of Downcodes reports: Recently, researchers have developed an AI model called DIAMOND (Diffusion for World Modelling), which can simulate the classic game "Counter-Strike: Global Offensive" (CS:GO) in a neural network, and in Running at 10 frames per second on an Nvidia RTX 3090 graphics card. Although the frame rate is not high, it is still a significant achievement in the field of AI simulation, especially considering that its training data is only 87 hours of game records, accounting for only 0.5% of the data required for similar projects.
Recently, researchers developed an AI model called DIAMOND (Diffusion for World Modeling), which can simulate the famous computer game "Counter-Strike: Global Offensive" (CS:GO) in a neural network.
This model runs on an Nvidia RTX3090 graphics card and is capable of reaching 10 frames per second . Although the frame rate is not high, this achievement is still impressive in the field of AI simulation.
The training data of DIAMOND is only 87 hours of CS:GO game records, which only accounts for 0.5% of the data required by similar projects such as GameNGen. Despite the small amount of data, this model is still able to simulate impressive scenes in the game.
DIAMOND first demonstrated its capabilities on Atari games, using a Transformer-based approach that treated the player's movements as "marks," like words in a sentence. By predicting these markers, the model can learn how to predict the player's next move based on previous actions.
Researcher Eloi Alonso demonstrated the model's capabilities on Twitter. In the video, players can be seen interacting with the simulated CS:GO environment through keyboard and mouse. Simulations include complex elements such as player interactions, weapon mechanics, and environmental physics. However, DIAMOND still has some significant flaws.
For example, the player can jump infinitely because the model does not take into account the Source engine's gravity or collision detection. Furthermore, once the player deviates from the path commonly used in the training data, the simulation breaks down completely.
The researchers believe that as the amount of data and computing power increases, the performance of the model will be further improved. They also believe that in the future it will be possible to develop AI models that can navigate complex real-world environments.
It is worth mentioning that DIAMOND’s CS:GO simulation is inspired by the GameNGen system jointly developed by Google Research, Google DeepMind and Tel Aviv University. This system can fully simulate the classic game DOOM at a speed of 20 frames per second on a single Google TPU chip. part of.
For developers interested in AI, the DIAMOND model is now open source on GitHub, and everyone is welcome to explore further.
Project entrance: https://diamond-wm.github.io/
Highlight:
- The AI model DIAMOND developed by researchers can simulate CS:GO and run on Nvidia RTX3090, reaching a speed of 10 frames per second.
- ? This model only used 87 hours of game data for training. Although the amount of data is small, it can still simulate complex game scenarios.
- ? DIAMOND has some serious limitations and vulnerabilities, but researchers believe model performance can be improved in the future by adding data and computing power.
The emergence of the DIAMOND model has brought new possibilities to the field of AI simulation games, and also provided valuable experience for the development of more complex AI models in the future. Although there are still some shortcomings, its potential cannot be underestimated. The editor of Downcodes looks forward to the further development and application of this model.