This repository contains the open-source implementation of MineAgent from the MineDojo paper (Fan et al., 2022). The authors described the agent’s architecture but did not release complete code—this repo fills that gap.
Tags / Keywords:
MineDojo, MineAgent, Minecraft AI, Open-ended RL, Hierarchical Agents, Skill Selector, Language-conditioned Policy, Embodied AI, Minecraft Reinforcement Learning, LLM-guided agent, Minecraft automation, Multimodal RL, Minecraft open-ended tasks, Self-imitation Learning, PPO, MineCLIP, Minecraft agent baseline, Computer Vision, Natural Language Conditioning, Action Planning, Imitation Learning, Minecraft Research
MineAgent is a hierarchical agent designed for open-ended tasks in Minecraft. It integrates:
- A language-conditioned policy
- A skill selector trained on human data
- A low-level controller that executes atomic actions
The agent can solve diverse tasks ranging from crafting tools to navigating terrain, using natural language instructions as input.
Linxi Fan, Guanzhi Wang, Yunfan Jiang, et al.
"MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledge"
NeurIPS 2022. [arXiv]