Wrapper for Gymnasium environments for reinforcement learning to work with POMDPs.jl. Includes options to get the observation space from pixels.
- Install PyCall.jl for julia
using Pkg; Pkg.add("PyCall")
- Install OpenAI gym in the same version of python used by
PyCall.jlusing Pkg; Pkg.add("Conda"); using Conda; Conda.add("gymnasium")
- (Optional) Install MuJoCo
- Install this package by opening julia and running
]add https://github.com/ancorso/POMDPGym.jl
Currently, the automatic installation using Conda.jl does not install the Atari environments of Gymnasium. To do this, install Atari environments in a custom Python environment manually and ask PyCall.jl to use it. To elaborate, create a new Python virtual environment and run
pip install gymnasium[classic-control] gymnasium[atari] pygame
pip install autorom
Then run the shell command AutoROM and accept the Atari ROM license. Now you can configure PyCall.jl to use your Python environment following the instructions here.
Optionally, you can also install MuJoCo.
Anthony Corso (acorso@stanford.edu)