An Open-Ended Embodied Agent with Large Language Models

Access to Voyager (Free)

Voyager Features Voyager is an innovative project that introduces the first LLM (Lifelong Learning Model)-powered embodied agent in the Minecraft environment. This agent autonomously explores the Minecraft world, acquires diverse skills, and makes novel discoveries without human intervention. Voyager comprises three key components that enable its lifelong learning capabilities. Key Features: Automatic Curriculum: Voyager autonomously maximizes exploration in Minecraft. Ever-Growing Skill Library: Voyager stores executable code for complex behaviors. Iterative Prompting Mechanism: Incorporates environment feedback, execution errors, and self-verification for program improvement. Interaction with GPT-4: Voyager utilizes GPT-4’s capabilities via blackbox queries. Temporally Extended, Interpretable, and Compositional Skills: Voyager’s skills compound rapidly and prevent catastrophic forgetting. Strong Lifelong Learning Capability: Exhibits exceptional proficiency in playing Minecraft and acquiring new skills. Generalization and Adaptation: Voyager can apply learned skills to solve novel tasks in new Minecraft worlds. Use Cases: AI researchers and developers interested in embodied lifelong learning. Minecraft enthusiasts seeking advanced AI agents to enhance their gameplay experience. Educational institutions exploring AI agents for virtual learning environments. AI enthusiasts interested in studying the capabilities of advanced language models in complex tasks. Voyager represents a significant step forward in embodied lifelong learning, specifically designed for the Minecraft environment.