Odyssey is a framework that empowers LLM-based Minecraft agents with open-world skills, featuring 40 primitive skills and 183 compositional skills, enabling AI to autonomously explore, learn, and execute diverse tasks in the Minecraft universe.
One-Minute Overview#
Odyssey is an innovative Minecraft agent framework built on Large Language Models, enabling AI agents to autonomously learn and explore in the open world of Minecraft. It features a comprehensive skill library with 40 primitive skills and 183 compositional skills, along with a dedicated LLaMA-3 model, allowing AI to not only complete traditional game tasks but also perform long-term planning and dynamic decision-making. For AI researchers and game developers, Odyssey provides a powerful platform to test and advance autonomous agent technologies.
Core Value: Breaks through limitations of traditional game tasks, enabling AI agents to autonomously explore and execute diverse tasks in open worlds.
Quick Start#
Installation Difficulty: Medium-High - Requires setting up multiple components including Python environment, Node.js, Minecraft server, and embedding models
# Python installation
cd Odyssey
pip install -e .
pip install -r requirements.txt
# Node.js installation
npm install -g yarn
cd Odyssey/odyssey/env/mineflayer
yarn install
cd Odyssey/odyssey/env/mineflayer/mineflayer-collectblock
npx tsc
Is this suitable for me?
- ✅ AI Research: Researchers testing and evaluating LLM agents in open-world environments
- ✅ Game Development: Developers exploring AI applications and interaction possibilities in games
- ❌ Casual Gaming: Not suitable for regular players who just want to experience AI in games
- ❌ Lightweight Projects: Requires strong technical background and computing resources to run
Core Capabilities#
1. Open-World Skill Library - Breaking Beyond Traditional Game Tasks#
- Contains 40 primitive skills and 183 compositional skills covering various scenarios from basic collection to complex combat Actual Value: AI agents are no longer limited to traditional tasks like "obtain diamond" but can execute a richer variety of game behaviors
2. Specialized Language Model - Enhanced Decision-Making#
- Based on LLaMA-3 model, fine-tuned on a dataset with 390k+ instruction entries Actual Value: Agents can better understand task descriptions and make reasonable decisions, reducing random behavior
3. Multi-Task Evaluation Benchmark - Comprehensive AI Testing#
- Includes long-term planning, dynamic-immediate planning, and autonomous exploration tasks Actual Value: Researchers can comprehensively evaluate AI agent performance across different scenarios, driving technological advancement
Tech Stack & Integration#
Development Languages: Python, Node.js Key Dependencies: LLaMA-3 model, Mineflayer (Minecraft interface), sentence-transformers (embedding model) Integration Method: API / SDK / Library
Ecosystem & Extension#
- Component Modules: Includes four main modules (LLM backend, MC crawler, model fine-tuning, and agent code) for easy extension and customization
- Data Collection Tools: Provides web crawler for collecting data from Minecraft Wiki, which researchers can modify to obtain domain-specific data
Maintenance Status#
- Development Activity: High - Frequent updates from June 2024 to April 2025 with new features and papers
- Recent Updates: April 2025, project accepted by IJCAI 2025 conference
- Community Response: Active - Provides detailed documentation and examples, with all data and model weights publicly available
Documentation & Learning Resources#
- Documentation Quality: Comprehensive - Includes detailed installation guides, configuration instructions, and task examples
- Official Documentation: Included in the GitHub repository
- Sample Code: Provides example code for various task scenarios that can be directly run and modified