Procgen¶
16 procedurally generated environments for testing generalisation.
- Install:
pip install -e ".[procgen]"- Paradigm:
Single-agent
- Stepping:
SINGLE_AGENT- Note:
Python 3.10 uses
procgen; Python 3.11+ usesprocgen-mirror
Environment |
Environment |
|---|---|
BigFish |
BossFight |
CaveFlyer |
Chaser |
Climber |
CoinRun |
DodgeBall |
FruitBot |
Heist |
Jumper |
Leaper |
Maze |
Miner |
Ninja |
Plunder |
StarPilot |
Citation¶
@inproceedings{cobbe2020procgen,
author = {Karl Cobbe and Christopher Hesse and Jacob Hilton and John Schulman},
title = {Leveraging Procedural Generation to Benchmark Reinforcement Learning},
booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
year = {2020},
}