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 3.3× more unique items, travels 2.3× longer distances, and unlocks key tech tree milestones up to 15.3× 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.
TIMELINE:
[00:00] - Intro
[00:48] - Voyager AI
[01:58] - Minecraft
[02:53] - Mineflayer API
[04:11] - GPT-4
[05:36] - Self-Improving
[07:51] - The Full Prompt Used
[11:18] - The Prompt for Creating Code
[17:58] - GPT-4 Decision Making
[19:00] - The Results
[22:16] - MineDojo (open source)
[23:35] - Implications for coding
The Paper: https://arxiv.org/abs/2305.16291