Voyager: An Open-Ended Embodied Agent with Large Language Models

By Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi (Jim) Fan, Anima Anandkumar. Trans. Mach. Learn. Res., 2023

1,998 citations recorded by the source corpus.

We introduce Voyager, the first LLM-powered embodied lifelong learning agent in Minecraft that continuously explores the world, acquires diverse skills, and makes novel discoveries without human intervention. Voyager consists of three key components: 1) an automatic curriculum that maximizes exploration, 2) an ever-growing skill library of executable code for storing and retrieving complex behaviors, and 3) a new iterative prompting mechanism that incorporates environment feedback, execution errors, and self-verification for program improvement. Voyager interacts with GPT-4 via blackbox queries, which bypasses the need for model parameter fine-tuning. The skills developed by Voyager are temporally extended, interpretable, and compositional, which compounds the agent's abilities rapidly and alleviates catastrophic forgetting. Empirically, Voyager shows strong in-context lifelong learning capability and exhibits exceptional proficiency in playing Minecraft. It obtains 3.3x more unique items, travels 2.3x longer distances, and unlocks key tech tree milestones up to 15.3x faster than prior SOTA. Voyager is able to utilize the learned skill library in a new Minecraft world to solve novel tasks from scratch, while other techniques struggle to generalize. We open-source our full codebase and prompts at https://voyager.minedojo.org/.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/f197bf0fc2f228483f6af3285000d54d8d97f9eb

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models (2023, neutral): “Leveraging pre-trained language models for embodied applications is an active area of research, where a large body of works focus on planning and reasoning with language models [9–11, 58, 31, 39, 59–72, 36, 73, 74].”
Mastering diverse control tasks through world models (2025, positive): “Voyager uses the substantially more abstract actions provided by the high-level MineFlayer bot scripting library, such as predefined behaviours for exploring the world until a resource is found and for automatically mining specified materials within a 32-m distance 47 .”
Large Language Models as Tool Makers (2023, methodology): “Our work aligns with the spirit of both Wang et al. (2023) and Qian et al. (2023) in the aim to let LLMs to generate their own tools for problem-solving.”
AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors in Agents (2023, neutral): “A brief introduction of the Voyager agent is provided here, and we refer the interested readers to Wang et al. (2023a) for a more detailed exposition.”
Aligning Cyber Space With Physical World: A Comprehensive Survey on Embodied AI (2024, neutral): “In addition, some approaches abstract past successful examples into a series of skills stored in a memory bank to consider during inference and improve planning success rates [179], [180], [181].”
AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation Framework (2023, background): “There are a few other specialized single-agent or multi-agent systems, such as Voyager [49] and Generative Agents [38], which we skip due to lower relevance.”
Mathematical discoveries from program search with large language models (2023, background): “68 the task was continuous exploration in the game of Minecraft.”
Making Large Language Models Better Reasoners with Alignment (2023, positive): “…Chung et al., 2022; Li et al., 2023a); 3) prompting : The prompting methods aim to improve the reasoning ability of LLMs by carefully designed prompting strategy without updating the model parameters (Wei et al., 2022; Wang et al., 2023c; Bi et al., 2023), which is very convenient and practical.”
SGLang: Efficient Execution of Structured Language Model Programs (2023, neutral): “Recent increases in the capabilities of LLMs have broadened their utility, enabling them to tackle a wider range of general tasks and act as autonomous agents [35, 6, 36, 52, 46].”
Foundation Models Defining a New Era in Vision: A Survey and Outlook (2025, neutral): “Furthermore, Mine-Dojo [158] gathers and trains agents on Minecraft data, while VOYAGER [159] is developed for lifelong learning based on LLM.”
TravelPlanner: A Benchmark for Real-World Planning with Language Agents (2024, methodology): “A new generation of language agents (Su, 2023; Sumers et al., 2023; Xie et al., 2023) powered by LLMs has emerged, characterized by their capability of using language as a vehicle for thought and communication.”
Large language models empowered agent-based modeling and simulation: a survey and perspectives (2023, positive): “Voyager Wang et al. (2024a) employ an automatic curriculum to aid the LLM agent in understanding the sequence of actions required to reach a goal.”
Large Language Models as Optimizers (2023, background): “…(Shinn et al., 2023; Madaan et al., 2023) and code generation performance (Chen et al., 2023e; Olausson et al., 2023; Shinn et al., 2023; Chen et al., 2023b), dialogue applications (Nair et al., 2023; Madaan et al., 2023; Yuan et al., 2023), and so on (Kim et al., 2023; Wang et al., 2023).”
Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces (2024, positive): “Prompting techniques improve the reasoning and problem-solving abilities for large models across diverse tasks [33, 35, 75, 84].”
Empowering Biomedical Discovery with AI Agents (2024, positive): “Through direct reasoning, agents can generate multiple threads of thought that could consider the best pathways, protein targets, and experiments that can be run to test the role of a candidate protein target (Figure 5c).”
Eureka: Human-Level Reward Design via Coding Large Language Models (2023, positive): “Their remarkable abilities in code writing, zero-shot generation, and in-context learning have previously enabled effective programmatic agents (Shinn et al., 2023; Wang et al., 2023a).”