Rethinking Decision Transformer via Hierarchical Reinforcement Learning

This is the official implementation for paper

Rethinking Decision Transformer via Hierarchical Reinforcement Learning

1. Create and activate conda environment

conda env create -f env.yaml

2. Additional installations

① Install the dependencies related to the mujoco-py environment. For more details, see https://github.com/openai/mujoco-py#install-mujoco

② Install D4RL. For more details, see https://github.com/Farama-Foundation/D4RL

3. Running

$ cd algorithms
# For V-ADT, first train IQL's Q/V, then train low-level DT
$ python iql.py
$ python V-ADT.py
# For G-ADT, first train HIQL's Q/V/goal, then train low-level DT
$ python hiql.py
$ python G-ADT.py

You can also modify the bash files to start the training. For the hyperparameters setting, please refer to the config folder and the appendix in the paper.

Please feel free to contact me if you have questions.