Sample Efficient Reinforcement Learning Through Human Demonstrations
This quick-start kit provides all of the resources needed to be successful in the MineRL Diamond Challenge.
** The full documentation for the dataset and the environments is found here!
- Submission, competition updates, and leaderboards are available via the competition homepage
note that submissions and leaderboards are not availabe yet
- Questions about getting started or rules of the competition should be directed to the competition forum or discord server
- Technical issues related to the code should be submitted through the MineRL GitHub page This repo may not be monitored!
The MineRL Competiton uses a custom distribution of Microsoft's Malmo Env. This environment is packaged in the minerl
package available via PyPI. The documentation can be found here
Ensure JDK-8 is installed and then simply (python3.5+)
pip3 install minerl --user
For a full guide please checkout the guide
minerl
uses OpenAI gym wrappers for the following environments with accompanying data:
MineRLTreechop-v0
MineRLNavigate-v0
MineRLNavigateDense-v0
MineRLNavigateExtreme-v0
MineRLNavigateExtremeDense-v0
MineRLObtainIronPickaxe-v0
MineRLObtainIronPickaxeDense-v0
MineRLObtainDiamond-v0
| All agents will be evaluated on this environmentMineRLObtainDiamondDense-v0
minerl
also currently includes a few debug environments for testing that lack any data:
MineRLNavigateDenseFixed-v0
MineRLObtainTest-v0
The MineRL Competition leverages a large-scale dataset of human demonstrations - MineRLv0. To ensure access during evaluation, a python api is provided to load demonstrations. Currently the data is almost 15GB, ensure ample space before downloading! To see how check out the quick start guide!
To get started with some baselines check out the chainerrl_baselines/
folder in this repository!
Submissions are coming soon!