1 min read

Google DeepMind has developed a new AI agent capable of playing 3D games

Google DeepMind has introduced a groundbreaking AI model named SIMA, short for ‘Scalable, Instructable, Multiworld Agent’, which is designed to revolutionize the way AI interacts with and plays 3D games. Unlike traditional AI agents that specialize in one game and focus solely on winning, SIMA is built to play multiple 3D games just like a human player. Moreover, it is equipped to understand and respond to natural language commands provided by gamers, making it a versatile virtual gaming partner.

This innovative AI agent is not confined to linear games with predetermined paths or stories; it is also capable of navigating and thriving in open-world gaming environments. Google DeepMind has collaborated with various game developers to train SIMA on a diverse range of video games, including popular titles like No Man’s Sky, Teardown, Valheim, and Goat Simulator 2. By partnering with these developers, DeepMind has ensured that SIMA is adept at understanding the dynamics of different gaming worlds and can execute tasks effectively based on the instructions it receives.

SIMA’s functionality is enabled by its ability to perceive and interpret various gaming environments and execute actions to achieve specified goals. It comprises a sophisticated model designed for accurate image-language mapping, coupled with a video model that predicts in-game events. Remarkably, SIMA does not require access to a game’s source code or specialized APIs; it only needs visual input from the game screen and straightforward, natural-language instructions from the user. Utilizing keyboard and mouse outputs, SIMA controls the game’s central character to carry out the provided instructions, mimicking the interaction methods commonly used by humans.


Currently, SIMA is proficient in performing over 600 fundamental gaming skills, encompassing tasks such as navigation, object interaction, and menu utilization. During a media briefing, DeepMind researcher Tim Harley emphasized that SIMA has been trained to execute commands within the game as directed by the user. Looking ahead, DeepMind aims to enhance SIMA’s capabilities to tackle more complex tasks requiring strategic planning and the execution of multiple sub-tasks.

The introduction of SIMA marks a significant advancement in the field of AI gaming agents, promising a more immersive and interactive gaming experience for players. By bridging the gap between AI technology and gaming, Google DeepMind has demonstrated the potential for AI to enhance various aspects of virtual environments. With further development and refinement, SIMA and similar AI models have the potential to transform the gaming landscape, offering players new levels of engagement and excitement.

Leave a Reply