gym
tool use environments.
$ pip install gym-tool-use
import gym_tool_use # import to register gym envs
env = gym.make("TrapTube-v0")
observation = env.reset()
action = env.action_space.sample()
observation_next, reward, done, info = env.step(action)
image = env.render(mode="rgb_array") # also supports mode="human"
The following environments are registered:
"TrapTube-v0"
(base task)"PerceptualTrapTube-v0"
"StructuralTrapTube-v0"
"SymbolicTrapTube-v0"
"PerceptualSymbolicTrapTube-v0"
"StructuralSymbolicTrapTube-v0"
"PerceptualStructuralTrapTube-v0"
"PerceptualStructuralSymbolicTrapTube-v0"
Baseline implementations here: https://github.com/fomorians/tool-use
Development is started with pipenv
.
$ pipenv install
$ pipenv shell
If you use this code in your work, please cite the following:
@ARTICLE{2019arXiv190702050W,
author = {{Wenke}, Sam and {Saunders}, Dan and {Qiu}, Mike and {Fleming}, Jim},
title = "{Reasoning and Generalization in RL: A Tool Use Perspective}",
journal = {arXiv e-prints},
keywords = {Computer Science - Neural and Evolutionary Computing, Computer Science - Artificial Intelligence, Computer Science - Machine Learning},
year = "2019",
month = "Jul",
eid = {arXiv:1907.02050},
pages = {arXiv:1907.02050},
archivePrefix = {arXiv},
eprint = {1907.02050},
primaryClass = {cs.NE},
adsurl = {https://ui.adsabs.harvard.edu/abs/2019arXiv190702050W},
adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}