Google has developed a versatile SIMA AI agent capable of playing games, collaborating, and executing actions based on your commands. Trained on nine video games, including No Man’s Sky, Teardown, Valheim, Hydroneer, and others, the SIMA AI agent can perform tasks in previously unseen games, making it a generalized multi-world agent.
The Google DeepMind team has introduced an AI agent known as SIMA (Scalable Instructable Multiworld Agent), capable of executing tasks across a diverse array of video games. What sets this development apart is that, for the first time, a general AI agent can follow user instructions to accomplish various tasks in a broad spectrum of virtual games, akin to human capabilities.
Unlike traditional approaches where AI agents are trained on individual video games confined to specific environments, SIMA is trained across multiple video games, rendering it a general, adaptable, and multi-world agent. Google DeepMind’s SIMA showcases its ability to comprehend unfamiliar gaming environments and carry out actions reminiscent of human behavior.
Additionally, Google DeepMind states that SIMA has been trained on images and natural language instructions. It can control gameplay using keyboard and mouse outputs. Acquiring nearly 600 skills, SIMA can perform a wide range of tasks in a virtual environment, from navigation to object interaction and opening maps, all within 10 seconds. The team aims to develop AI agents capable of planning and executing actions.
Read Also:
OpenAI Launches Sora: A Revolutionary Text-to-Video AI Model
When compared to specialized agents trained on individual games, SIMA surpasses them all in performance. To comprehend how the agent operates, you can delve into the technical report on SIMA authored by the Google DeepMind team. While it operates as an AI agent executing tasks in a virtual environment, the utilization of such generalized AI agents holds significant potential in robotics, signifying a paradigm shift in the real world.