Voyager is an embodied lifelong learning agent in Minecraft powered by a large language model (LLM) called GPT-4. It explores the Minecraft world, acquires skills, and makes novel discoveries without human intervention. Voyager consists of an automatic curriculum for exploration, a skill library for storing complex behaviors, and an iterative prompting mechanism for program improvement.
- Voyager discovers new Minecraft items and skills continually through self-driven exploration, outperforming baseline methods.
- 🚀 Automatic Curriculum: Voyager utilizes an automatic curriculum to maximize exploration and guide its learning process.
- 📚 Skill Library: Voyager has an ever-growing skill library that stores and retrieves complex behaviors, allowing it to accumulate and reuse mastered skills.
- 🔄 Iterative Prompting Mechanism: Voyager incorporates environment feedback, execution errors, and self-verification to improve its programs through an iterative prompting mechanism.
- 🌍 In-Context Lifelong Learning: Voyager demonstrates strong in-context lifelong learning capability, rapidly compounding its abilities and alleviating catastrophic forgetting.
- 🎮 Minecraft Proficiency: Voyager performs exceptionally well in playing Minecraft, obtaining more unique items, traveling longer distances, and unlocking key tech tree milestones faster than previous state-of-the-art methods.
- 🧠 Compositional Skills: The skills developed by Voyager are temporally extended, interpretable, and compositional, enabling the agent to solve novel tasks from scratch in new Minecraft worlds.
- 🤖 LLM Interactions: Voyager interacts with GPT-4, a blackbox LLM, through blackbox queries, eliminating the need for fine-tuning the model’s parameters.