danijar/crafter

Benchmarking the Spectrum of Agent Capabilities

64
/ 100
Established

A 2D procedurally-generated survival environment with 22 semantic achievements (foraging, crafting, combat, shelter) evaluated across both reward-based and unsupervised agents. Follows the OpenAI Gym interface with 64×64 RGB observations and 17 discrete actions, enabling comprehensive agent benchmarking within a single environment rather than across separate tasks. Evaluation uses geometric mean scoring of achievement success rates over a 1M-step budget, accommodating RL agents, exploration methods, and external knowledge approaches.

527 stars and 2,503 monthly downloads. No commits in the last 6 months. Available on PyPI.

Stale 6m No Dependents
Maintenance 0 / 25
Adoption 18 / 25
Maturity 25 / 25
Community 21 / 25

How are scores calculated?

Stars

527

Forks

87

Language

Python

License

MIT

Last pushed

Jan 23, 2024

Monthly downloads

2,503

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/agents/danijar/crafter"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.