miHoYo Introduces Gamer: A Generalist Vision-Language-Action Game Agent
The miHoYo Honkai AI R&D team has introduced 'Gamer', a generalist Vision-Language-Action foundation model designed for long-horizon, real-time video game interaction. Unlike traditional narrow or game-specific agents, Gamer interacts with virtual worlds through the exact same interfaces that human players use. This development represents a major step forward in creating versatile AI agents capable of handling complex, dynamic 2D and 3D gaming environments. It highlights the growing trend of applying foundation models and reinforcement learning to build interactive digital entities that mirror human gameplay. Gamer integrates core capabilities such as perception, memory, and multi-step reasoning through techniques like continued pre-training, sparse-thinking instruction tuning, and multi-turn reinforcement learning. It is developed by miHoYo, a company renowned for blockbuster gacha titles like Genshin Impact and Honkai: Star Rail.
## BACKGROUND
Vision-Language-Action foundation models combine computer vision, natural language processing, and robotic or software actions to understand and act within a digital environment. Reinforcement learning is a machine learning method where agents learn to make decisions by receiving rewards or penalties for their actions in an environment.