Voyager
Visit Website
About
Voyager is the first Large Language Model (LLM)-powered embodied lifelong learning system designed to autonomously explore, master a wide range of skills, and make new discoveries in Minecraft without human intervention. It comprises three key components: an automatic curriculum that maximizes exploration, a skill library 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 fine-tuning. The skills developed are temporally extended, interpretable, and compositional, rapidly enhancing the system's capabilities and mitigating catastrophic forgetting.
Features
- Automatic curriculum generation for open-ended exploration.
- Ever-growing skill library for storing and retrieving complex behaviors.
- Iterative prompting mechanism incorporating environment feedback and self-verification.
- Interaction with GPT-4 via blackbox queries without model fine-tuning.
- Temporally extended, interpretable, and compositional skill development.