game-playing agents

Autonomous software systems that use reinforcement learning and natural language to navigate, strategize, and master complex virtual environments.

Game-playing agents have evolved from single-purpose algorithms like AlphaStar (which mastered StarCraft II) into highly adaptable generalists. Modern systems like Google DeepMind's SIMA (Scalable Instructable Multiworld Agent) do not rely on game-specific source code or custom APIs; instead, they interact with virtual worlds through standard keyboard-and-mouse inputs and real-time visual processing. By pairing deep reinforcement learning with natural-language understanding, these agents can execute hundreds of tasks across diverse 3D environments (such as No Man's Sky and Valheim) based on simple verbal instructions. This rapid progress in virtual spatial-temporal reasoning serves as a critical sandbox for training the next generation of real-world robotics and embodied physical AI.

https://deepmind.google/discover/blog/sima-generalist-ai-agent-for-3d-virtual-environments/