GitHub

Voyager is the first LLM-powered embodied lifelong learning agent in Minecraft. It continuously explores the world, acquires diverse skills, and makes novel discoveries without human intervention. The system consists of three key components: an automatic curriculum that maximizes exploration, an ever-growing skill library of executable code for storing and retrieving complex behaviors, and an iterative prompting mechanism that incorporates environment feedback, execution errors, and self-verification for program improvement. Voyager interacts with GPT-4 via blackbox queries, bypassing the need for model parameter fine-tuning. Its skills are temporally extended, interpretable, and compositional, compounding the agent's abilities rapidly while alleviating catastrophic forgetting. Empirically, Voyager obtains 3.3× more unique items, travels 2.3× longer distances, and unlocks key tech tree milestones up to 15.3× faster than prior state-of-the-art methods. It can utilize its learned skill library in new Minecraft worlds to solve novel tasks from scratch, demonstrating strong in-context lifelong learning capability.