Pre-requisites:

• miniconda

Clone this repository with

cd ~
git clone https://github.com/rml-unist/PCEG-code.git
cd PCEG-code

Download IsaacGym Preview 4 and extract it under ./deps/isaacgym

mv ~/Downloads/IsaacGym_Preview_4_Package.tar.gz ~/PCEG-code/deps/
cd ~/PCEG-code/deps
tar -xvf IsaacGym_Preview_4_Package.tar.gz

Run the bash setup script to install everything.

cd ~/PCEG-code
bash setup.sh

To try out the inference, first download the data and the trained models.

conda activate mpd

gdown --id 1mmJAFg6M2I1OozZcyueKp_AP0HHkCq2k
tar -xvf data_trajectories.tar.gz
gdown --id 1I66PJ5QudCqIZ2Xy4P8e-iRBA8-e2zO1
tar -xvf data_trained_models.tar.gz

After downloading, please change the below variables

TRAINED_MODELS_DIR (scripts/inference/inference.py)
ROBOT_ASSET_DIR (scripts/inference/inference.py : .../isaacgym/asset)
data_dir (mpd/datasets/trajectories.py)

Run the inference script

cd scripts/inference
python inference.py

Comment out the model-id variable in scripts/inference/inference.py to try out different models

model_id: str = 'EnvDense2D-RobotPointMass'
model_id: str = 'EnvNarrowPassageDense2D-RobotPointMass'
model_id: str = 'EnvSimple2D-RobotPointMass'
model_id: str = 'EnvSpheres3D-RobotPanda'

The results will be saved under ./[model_id]/.

The most of this repository is from

Carvalho, J.; Le, A.T.; Baierl, M.; Koert, D.; Peters, J. (2023). Motion Planning Diffusion: Learning and Planning of Robot Motions with Diffusion Models, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).