A Practical Guide for Incorporating Symmetry in Diffusion Policy

Project Website | Paper
Dian Wang¹, Boce Hu², Shuran Song¹, Robin Walters², Robert Platt^2
1Stanford Univeristy, ²Northeastern Univeristy
NeurIPS 2025

Installation

1. Install the following apt packages for mujoco:

   sudo apt install -y libosmesa6-dev libgl1-mesa-glx libglfw3 patchelf

2. Install gfortran (dependancy for escnn)

   sudo apt install -y gfortran

3. Install Mambaforge (recommended) or Anaconda

4. Install environment:

   mamba env create -f conda_environment.yaml
   conda activate symindp

   or:

   conda env create -f conda_environment.yaml
   conda activate symindp

5. Force reinstall lie-learn (due to a known issue)

   pip uninstall lie-learn
   pip install git+https://github.com/AMLab-Amsterdam/lie_learn@07469085ac0fd4550fd26ff61cb10bb1e92cead1

6. Install robosuite:
   Notice that this fork of robosuite has the large FOV in-hand camera. To use the standard FOV camera, do git clone https://github.com/ARISE-Initiative/robosuite.git -b b9d8d3de5e3dfd1724f4a0e6555246c460407daa instead.

   cd ..
   git clone https://github.com/BoceHu/robosuite.git
   cd robosuite
   pip install -e .
   cd ..

7. Install robomimic:

   git clone https://github.com/pointW/robomimic.git
   cd robomimic
   pip install -e .
   cd ..

8. Install robosuite-task-zoo:

   git clone https://github.com/pointW/robosuite-task-zoo.git
   cd robosuite-task-zoo
   pip install -e .
   cd ..

9. Install mimicgen:

   git clone https://github.com/NVlabs/mimicgen_environments.git
   cd mimicgen_environments
   git checkout 45db4b35a5a79e82ca8a70ce1321f855498ca82c
   pip install -e .
   cd ../sym_in_dp

10. Make sure mujoco version is 2.3.2 (required by mimicgen)

    pip list | grep mujoco

Dataset

Download Dataset

Download dataset from MimicGen's hugging face: https://huggingface.co/datasets/amandlek/mimicgen_datasets/tree/main/core
Make sure the dataset is kept under /path/to/sym_in_dp/data/robomimic/datasets/[dataset]/[dataset].hdf5

Generating Large FOV Observation

# Template
python sym_in_dp/scripts/dataset_states_to_obs.py --input data/robomimic/datasets/[dataset]/[dataset].hdf5 --output data/robomimic/datasets/[dataset]/[dataset]_fisheye.hdf5 --num_workers=[n_worker]
# Replace [dataset] and [n_worker] with your choices.
# E.g., for stack_d1
python sym_in_dp/scripts/dataset_states_to_obs.py --input data/robomimic/datasets/stack_d1/stack_d1.hdf5 --output data/robomimic/datasets/stack_d1/stack_d1_fisheye.hdf5 --num_workers=24

Convert Action Space in Dataset

The downloaded dataset has a relative action space. To train with absolute action space, the dataset needs to be converted accordingly

# Template
python sym_in_dp/scripts/robomimic_dataset_conversion.py -i data/robomimic/datasets/[dataset]/[dataset].hdf5 -o data/robomimic/datasets/[dataset]/[dataset]_abs.hdf5 -n [n_worker]
# Replace [dataset] and [n_worker] with your choices.
# E.g., convert stack_d1_fisheye with 12 workers
python sym_in_dp/scripts/robomimic_dataset_conversion.py -i data/robomimic/datasets/stack_d1/stack_d1_fisheye.hdf5 -o data/robomimic/datasets/stack_d1/stack_d1_fisheye_abs.hdf5 -n 12

Training Rel Traj CNN Enc

python train.py --config-name=train_diffusion_unet_rel_traj task_name=stack_d1

To train in other tasks, replace stack_d1 with stack_three_d1, square_d2, threading_d2, coffee_d2, three_piece_assembly_d2, hammer_cleanup_d1, mug_cleanup_d1, kitchen_d1, nut_assembly_d0, pick_place_d0, coffee_preparation_d1. Notice that the corresponding dataset should be downloaded already.

Training Rel Traj Equi Enc

python train.py --config-name=train_diffusion_unet_equi_enc_rel_traj task_name=stack_d1

Training Rel Traj Pretrain

python train.py --config-name=train_diffusion_unet_pretrained_res18_rel_traj task_name=stack_d1

Training Rel Traj Pretrain + FA

python train.py --config-name=train_diffusion_unet_pretrained_res18_fa_rel_traj task_name=stack_d1