README.md

LatentSpacePolicies_for_HierarchicalRL

Pytorch implementation of "Latent Space Policies for Hierarchical Reinforcement Learning"

Install required packages with pip3

pip3 install -r requirements.txt

Try out

Train agent in "HumanoidBulletEnv" in PyBullet library.
A dedicated configuration file for this environment is in configs directory.

python3 main.py --config configs/config_lsphrl_humanoid.toml --save-dir results/test_humanoid

A model file is saved in results/test_humanoid/checkpoints/model.pth. Load this file to run the trained policy.

python3 main.py --config configs/config_lsphrl_humanoid.toml -m results/test_humanoid/checkpoints/model.pth -e -r

Options

--config              Config file path
--save-dir            Save directory
--visualize-interval  Interval to draw graphs of metrics.
--device              Device for computation.
-e, --eval            Run model evaluation.
-m, --model-filepath  Path to trained model for evaluation.
-r, --render          Render agent behavior during evaluation.

Implementation notes

hoge

Evaluations

foo

HumanoidBulletEnv