NetHack Learning Environment
Not yet importedA standardized interface to a procedurally generated, partially observed, system-rich, and extremely long-horizon game.
1 task previewagent · sandboxedMITversion upstream
Complex import. Requires long-lived seeded sandboxes, NLE observations and actions, progression metrics, and replayable episodes.
What it measures
NLE exposes NetHack tasks and observations for agents. Useful evaluation should report progression and task-level rewards rather than treating full ascension as the only success signal.
The environment is a poor first integration: episodes are long, failures are difficult to explain, and reproducible replay requires careful seed, action, and state capture.
not yet imported
Not yet imported. Build on a proven MiniHack adapter and publish task-level progression rather than one ascension score.
Tasks
1 previewed| id | task | runtime |
|---|---|---|
| nle/score | Score progressionMaximize progress and score through a long-horizon seeded episode.tools: NLE observation, game action | sandbox |
Tools
2 requested| tool | description | kind | required |
|---|---|---|---|
| NLE observation | Returns structured and language-wrapped NetHack state. | sandbox | yes |
| game action | Submits discrete actions to the active episode. | sandbox | yes |