All Tools
M
AgentsFreeOpen Source
MINERL
Minecraft environment and dataset for sample-efficient RL
CC-BY-NC-SA-4.0
ABOUT
Open-world pixel control is hard to study without a shared game and human demos. MineRL wraps Minecraft as a Gym env and ships competition datasets so agents can learn from pixels plus demonstrations instead of exploring from scratch.
INSTALL
pip install minerlINTEGRATION GUIDE
1. Train a sample-efficient RL agent to obtain diamonds in Minecraft
2. Imitate human MineRL-v0 demonstrations before online fine-tuning
3. Benchmark BASALT or Diamond competition tasks in a Gym loop
4. Collect pixel observations and discrete actions for hierarchical agents
TAGS
pythonminecraftreinforcement-learningimitation-learninggymdataset