MiniHack

Sandbox RL environments built on the NetHack Learning Environment (NLE).

Install:

pip install -e ".[nethack]"

Paradigm:

Single-agent

Stepping:

SINGLE_AGENT

System:

Requires build-essential, cmake, flex, bison

Category	Environments
Navigation	Room-5×5/15×15, Corridor-R2/R3/R5, MazeWalk-9×9/15×15/45×19, River, River-Narrow
Skills	Eat, Wear, Wield, Zap, Read, Quaff, PutOn, LavaCross, WoD-Easy/Medium/Hard
Exploration	ExploreMaze-Easy/Hard, HideNSeek, Memento-F2/F4

Citation

@inproceedings{samvelyan2021minihack,
  author       = {Mikayel Samvelyan and Robert Kirk and Vitaly Kurin and Jack Parker-Holder and Minqi Jiang and Eric Hambro and Fabio Petroni and Heinrich K{\"u}ttler and Edward Grefenstette and Tim Rockt{\"a}schel},
  title        = {MiniHack the Planet: A Sandbox for Open-Ended Reinforcement Learning Research},
  booktitle    = {Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks},
  year         = {2021},
}