Jakarta, INTI – Google DeepMind gained global attention in November, 16, 2025 when it introduced SIMA 2, an AI agent created to overcome complex challenges in virtual environments. The launch responds to growing demands for AI that can understand user intentions, adapt to new situations, and support the development of multi-purpose robots. Released as a research preview for academics and developers, SIMA 2 demonstrates early capabilities as a multi-world learning agent. It can learn from various 3D games and navigate unfamiliar settings. Google believes these virtual worlds provide an effective training ground before the technology is applied in real-world scenarios.
An AI Agent Capable of Learning Across Multiple Virtual Worlds
SIMA 2, or Scalable Instructable Multiworld Agent 2, is designed not only to follow instructions but also to deeply understand user intentions. The agent can navigate a range of 3D video games such as No Man’s Sky, Valheim, and Goat Simulator 3, including environments it has never encountered before. This adaptive ability marks an important step toward developing a general learning agent capable of handling diverse and dynamic scenarios.
Combining the Power of the Gemini Large Language Model
In its latest version, Google has integrated the Gemini large language model as the core processing engine of SIMA 2. This integration enables the agent not only to execute instructions directly but also to formulate strategies based on a deeper understanding of its environment. DeepMind explains that virtual games serve as a training ground to simulate complex situations from navigation to multi-step task completion before the technology is applied to physical robots.
Entering the Era of Artificial General Intelligence (AGI)
DeepMind views the development of SIMA 2 as a significant milestone in the global race toward Artificial General Intelligence (AGI). They believe that an agent capable of adapting to new environments without specific training is essential for building future versatile robots. However, Google emphasizes that SIMA 2 is not intended as a consumer or gaming product at this stage, but rather as a research experiment aimed at testing the limits of general learning agents. DeepMind also underscores that the primary goal of this development is to understand how AI agents can respond to new and unpredictable situations while paving the way for more intelligent and autonomous robotic systems.
Conclusion:
Through SIMA 2, Google demonstrates an ambitious step toward creating AI that can think and adapt like humans. Although still in the research phase, this technology has the potential to become the foundation for versatile robots capable of operating across both physical and digital worlds.
Read More: China’s AI Developer Community Accelerates Innovation and Expands Global Impact