This repository contains code for our work on 'Dynamic Inference on Graphs using Structured Transition Models' accepted for publication at the International Conference on Intelligent Robots and Systems (IROS) 2022.
Link to the paper: https://arxiv.org/abs/2209.15132
git clone git@github.com:SaumyaSaxena/Dynamic_GNN_structured_models.git
virtualenv -p python3.6 venv_gnn
source venv_gnn/bin/activate
cd Dynamic_GNN_structured_models/
pip install -e .
sudo apt-get install build-essential python-dev swig python-pygame
pip install git+https://github.com/pybox2d/pybox2d
pip install gym[box2d]
# Check if installed properly
python -c "import Box2D"
git clone git@github.com:iamlab-cmu/isaacgym.git
pip install -e isaacgym/python/
git clone git@github.com:iamlab-cmu/isaacgym-utils.git
pip install -e isaacgym-utils/[all]
pip install torch==1.9.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch-scatter==2.0.8 -f https://data.pyg.org/whl/torch-1.9.1+cu111.html
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.9.1+cu111.html
pip install torch-cluster -f https://data.pyg.org/whl/torch-1.9.1+cu111.html
pip install torch-spline-conv -f https://data.pyg.org/whl/torch-1.9.1+cu111.html
pip install torch-geometric==1.7.2 -f https://data.pyg.org/whl/torch-1.9.1+cu111.html
pip install pytorch_lightning
# Check if installed properly
python -c "import torch_geometric"
To collect data for training the forward model run:
python scripts/collect_skill_data_box2D_envs.py
Use the environment config file cfg/envs/block_grasp_box2D_env.yaml to vary the number of blocks in the scene by changing the lists: cfg['env_props']['blocks']['positions'] and cfg['env_props']['blocks']['velocities'].
To train the model run:
HYDRA_FULL_ERROR=1 python scripts/train_forward_model.py
Table below summarizes the our proposed model and ablation studies mentioned in the paper and respective datasets for training:
| Model | Model name | Dataset |
|---|---|---|
| GIM_Temp (proposed model) | EPDTemporalObsLinearModelwithModesReactive | Box2DEnvPickup2ObjsTemporalDataset |
| No-GIM | EPDLinearObsSpringMassDamperModelwithModes | Box2DEnvPickup2ObjsDataset |
| No-GIM-Aug | EPDLinearObsSpringMassDamperModel | Box2DEnvPickup1Obj1DistractorPickup2ObjsDatasetMixed |
| GIM_Non-Temp | EPDLinearObsSpringMassDamperModelwithModes | Box2DEnvPickup2ObjsDataset |
Use the training configuration file cfg/train/train_forward_model.yaml to choose the dataset and model to train.