Piyushi-0/gradient

Code base for NeurIPS 2022 paper Curriculum Reinforcement Learning using Optimal Transport via Gradual Domain Adaptation.

GRADIENT: Curriculum Reinforcement Learning using Optimal Transport via Gradual Domain Adaptation

Paper

Huang P, Xu M, Zhu J, Shi L, Fang F, Zhao D. Curriculum reinforcement learning using optimal transport via gradual domain adaptation. Advances in Neural Information Processing Systems. 2022 Dec 6;35:10656-70. https://arxiv.org/abs/2210.10195

Install dependencies

conda create --name gradient python=3.8.12
pip install -r requirements.txt

cd envs/gym && pip install -e . 

cd envs/mujoco-maze && pip install -e . 

Environments:

• Environments are modified from Mujoco_maze (https://github.com/kngwyu/mujoco-maze) and gym (https://github.com/openai/gym).

Code Usage

python  run_maze_continuous.py --curriculum gradient --interp_metric encoding --num_stage 5 --reward_threshold 0.5
python  run_maze_continuous.py --curriculum gradient --interp_metric l2 --num_stage 5 --reward_threshold 0.5