About Voyager
Voyager is an autonomous AI agent for Minecraft that leverages large language models (LLMs), specifically GPT-4, to achieve lifelong learning in an open-ended environment. It can independently explore, plan, and execute complex skills through an automatic curriculum, a compositional skill library, and iterative prompting, all without human intervention. Voyager is designed for research in embodied AI, reinforcement learning, and autonomous skill acquisition, excelling in self-driven exploration, skill mastery, and transfer to new tasks.
Latest News about Voyager
Telefónica strengthens its leadership in autonomous networks at DTW Ignite 2026 and receives global recognition from TM Forum
Telefónica consolidates its position at the forefront of the global telco ecosystem with advances in the roll-out of Level 4 autonomous networks.
AI agent 'Voyager' that automatically plays Minecraft while iteratively learning with GPT-4, a large-scale language model, appears
Minecraft, the best-selling game in the world, continues to explore and acquire more materials, upgrade your tools and weapons, and expand the world you can explore. Developed by a joint research team such as NVIDIA, California Institute of Technology, and Stanford University, `` Voyager '' is a lifelong learning agent that learns and captures Minecraft without human intervention, and is OpenAI's large-scale language model (LLM). It is equipped with GPT-4 which is. Voyager | An Open-Ended Embodied Agent with Large Language Models https://voyager.minedojo.org/ GitHub - MineDojo/Voyager: An Open-Ended Embodied Agent with Large Language Models https://github.com/MineDojo/Voyager In Minecraft, the world is created based on the seed value, so as long as the seed value is random, the generated world will change greatly each time. Therefore, what is important in playing Minecraft is 'the ability to respond flexibly to a given situation'. It is also important to evolve the processing technology for materials that affect the performance of tools and weapons, such as wood, stone, iron, and diamond. Voyager consists of three components: 'automatic curriculum to maximize exploration', 'skill library for storing and retrieving complex actions', 'environmental feedback, execution errors, and self-verification for program improvement'. Learn iteratively. Voyager writes a program in GPT-4 to achieve a goal, and refines the program in GPT-4 based on feedback and errors from the game environment. By doing this, Vyager will gradually build up its skill library, prioritizing successful programs and building more complex skills. Click the image below to see Voyager actually mining gold ore. You can also watch the scene where Voyager fights Enderman by clicking the image below. The graph below shows the evolution of Voyager's skill tree. Orange is Voyager, blue is Voyager without a skill library, and green, red, and purple are language model-based minecraft agents other than Voyager, ReAct, Reflection, and AutoGPT, respectively. The vertical axis is the number of items found, and the horizontal axis is the number of code generation iterations. Looking at this, you can see that Voyager is making diamond weapons with overwhelming speed. In addition, since Voyager is operating on a text basis at the time of writing the article, it is not aware of what is actually happening on Minecraft. However, when the research team gave Voygaer visual feedback in the early stages of the experiment, it became possible to learn the construction of houses and Nether portals. Click on the thumbnail below to see how the nether portal is actually built.
Resources
Product Website
Visit Voyager's official website for product details and getting started.