This research aims to build general agents for open-world environments, leveraging large models to integrate perception, planning, learning, and control into a unified agent framework. The goal is to enable embodied intelligence that interacts with the real world, and to explore foundational agent models toward Artificial General Intelligence (AGI).

Representative Work:

Generalist Agent in the Open World of Minecraft

• Ghost in the Minecraft: Generally Capable Agents for Open-World Environments via Large Language Models with Text-based Knowledge and Memory

[Project page]