LIBERTY:Efficient Potential-Based Exploration in Reinforcement Learning Using the Inverse Dynamic Bisimulation Metric
This is the official implementation of NeurIPS 2023 paper - [Efficient Potential-Based Exploration in Reinforcement Learning Using the Inverse Dynamic Bisimulation Metric]
- cuda-10.0
- pytorch==1.2.0
- gym==0.12.5
- gym[atari]
- gym_super_mario_bros
- mujoco-py==1.50.1.56
- mpi4py
- pillow
- tqdm
- Please install the required packages in the above list.
- Install
utils
:
pip install -e .
Please enter ppo
folder to run LIBERTY
cd ppo/
python -u train.py --env-name='HalfCheetah-v2' --cuda (if cuda is available) --log-dir='logs' --seed=777