Skip to content

erdiphd/DiCuRL

Repository files navigation

OUTPACE

For DiCuRL, we utilized the original implementation of OUTPACE and augmented this codebase with our diffusion model for curriculum goal generation.

Setup Instructions

  1. Create a conda environment:
conda env create -f outpace.yml
conda activate outpace
  1. Add the necessary paths:
conda develop meta-nml
  1. Install subfolder dependencies:
cd meta-nml && pip install -r requirements.txt
cd ..
chmod +x install.sh
./install.sh
  1. Install pytorch (use tested on pytorch 1.12.1 with CUDA 11.3)

  2. Set config_path: see config/paths/template.yaml

  3. To run robot arm environment install metaworld:

pip install git+https://github.com/rlworkgroup/metaworld.git@3ced29c8cee6445386eba32e92870d664ad5e6e3#egg=metaworld

Usage

Training and Evaluation

PointUMaze-v0

CUDA_VISIBLE_DEVICES=0 python outpace_train.py env=PointUMaze-v0 aim_disc_replay_buffer_capacity=10000 save_buffer=true adam_eps=0.01

PointNMaze-v0

CUDA_VISIBLE_DEVICES=0 python outpace_train.py env=PointNMaze-v0 aim_disc_replay_buffer_capacity=10000 adam_eps=0.01

PointSpiralMaze-v0

CUDA_VISIBLE_DEVICES=0 python outpace_train.py env=PointSpiralMaze-v0 aim_disc_replay_buffer_capacity=20000 save_buffer=true aim_discriminator_cfg.lambda_coef=50

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages