VoxAct-B: Voxel-Based Acting and Stabilizing Policy for Bimanual Manipulation

[Project website] [Paper]

This project is a PyTorch implementation of VoxAct-B: Voxel-Based Acting and Stabilizing Policy for Bimanual Manipulation, published in CoRL 2024.

Authors: I-Chun Arthur Liu, Sicheng He, Daniel Seita*, Gaurav S. Sukhatme* (* equal advising).

Bimanual manipulation is critical to many robotics applications. In contrast to single-arm manipulation, bimanual manipulation tasks are challenging due to higher-dimensional action spaces. Prior works leverage large amounts of data and primitive actions to address this problem, but may suffer from sample inefficiency and limited generalization across various tasks. To this end, we propose VoxAct-B, a language-conditioned, voxel-based method that leverages Vision Language Models (VLMs) to prioritize key regions within the scene and reconstruct a voxel grid. We provide this voxel grid to our bimanual manipulation policy to learn acting and stabilizing actions. This approach enables more efficient policy learning from voxels and is generalizable to different tasks. In simulation, we show that VoxAct-B outperforms strong baselines on fine-grained bimanual manipulation tasks. Furthermore, we demonstrate VoxAct-B on real-world 𝙾𝚙𝚎𝚗 𝙳𝚛𝚊𝚠𝚎𝚛 and 𝙾𝚙𝚎𝚗 𝙹𝚊𝚛 tasks using two UR5s.

Installation

Prerequisites

VoxAct-B is built-off the PerAct repository by Shridhar et al. The prerequisites are the same as PerAct.

1. Environment

# setup a virtualenv with whichever package manager you prefer
conda create -n voxactb python=3.8
conda activate voxactb
pip install --upgrade pip==24.0

2. PyRep and Coppelia Simulator

Follow instructions from the official PyRep repo; reproduced here for convenience:

PyRep requires version 4.1 of CoppeliaSim. Download:

• Ubuntu 16.04
• Ubuntu 18.04
• Ubuntu 20.04
• Ubuntu 22.04

Add the following to your ~/.bashrc file: (NOTE: the 'EDIT ME' in the first line)

export COPPELIASIM_ROOT=<EDIT ME>/PATH/TO/COPPELIASIM/INSTALL/DIR
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$COPPELIASIM_ROOT
export QT_QPA_PLATFORM_PLUGIN_PATH=$COPPELIASIM_ROOT

Remember to source your bashrc (source ~/.bashrc) or zshrc (source ~/.zshrc) after this.

Warning: CoppeliaSim might cause conflicts with ROS workspaces.

Finally install the python library:

cd PyRep
pip install -r requirements.txt
pip install .

You should be good to go! You could try running one of the examples in the examples/ folder.

If you encounter errors, please use the PyRep issue tracker.

3. RLBench

cd ../RLBench
pip install -r requirements.txt
python setup.py develop

For running in headless mode, tasks setups, and other issues, please refer to the official repo.

4. YARR

cd ../YARR
pip install -r requirements.txt
python setup.py develop

Common Issues:

pkg_resources.extern.packaging.requirements.InvalidRequirement: .* suffix can only be used with `==` or `!=` operators
    PyYAML (>=5.1.*)

# Solution
pip install setuptools==66

5. [Optional] urx for real-robot experiments

cd urx
python setup.py develop

PerAct Repo

Install:

cd ../peract
pip install git+https://github.com/openai/CLIP.git
pip install -r requirements.txt

export PERACT_ROOT=$(pwd)  # mostly used as a reference point for tutorials
python setup.py develop

pip install transformers==4.40.0
pip install git+https://github.com/facebookresearch/segment-anything.git
pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116 # feel free to ignore "ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts."

Create data folder:

# make sure you're inside the `voxactb/peract` folder
mkdir data
cd data

Download this zip file, unzip it, and place clip_rn50.pth and segment_anything_vit_h.pth inside the voxactb/peract/data folder.

VoxPoser

Install:

# make sure you're inside the `voxactb/peract/voxposer` folder
pip install -r requirements.txt

[Optional] Running on a Headless Computer

Install:

sudo apt-get install mesa-utils x11-xserver-utils xvfb dbus-x11 x11-utils libxkbcommon-x11-0 

Export display variable:

export DISPLAY=:0.0

For each training/evaluation script, insert the following line before the python command:

xvfb-run -a --server-num=611 python <script>

Gotchas

OpenGL Errors

GL errors are probably being caused by the PyRender voxel visualizer. See this issue for reference. You might have to set the following environment variables depending on your setup:

export DISPLAY=:0
export MESA_GL_VERSION_OVERRIDE=4.1
export PYOPENGL_PLATFORM=egl

Unpickling Error

If you see _pickle.UnpicklingError: invalid load key, '\x9e', maybe one of the replay pickle files got corrupted when quitting the training script. Try deleting files in replay.path and restarting training.

How to run experiments

Datasets are hosted on Hugging Face.

• You can download the train/val/test datasets from here (place the train, val, and test folders in the voxactb/peract/data folder) or generate the datasets by following these instructions.

• To reproduce our results, you can download the checkpoints and place them in the voxactb/peract/log folder.

Go to voxactb/peract/scripts.

Open Jar

# training
./train_open_jar_ours_vlm_10_demos_v2_11_acting.sh
./train_open_jar_ours_vlm_10_demos_v2_11_stabilizing.sh

# validation/test
./eval_open_jar_ours_vlm_10_demos_v2_vlm_11.sh

Open Drawer

# training
./train_open_drawer_ours_vlm_10_demos_v2_11_acting.sh
./train_open_drawer_ours_vlm_10_demos_v2_11_stabilizing.sh

# validation/test
./eval_open_drawer_ours_vlm_10_demos_v2_vlm_11.sh

Put Item in Drawer

# training
./train_put_item_in_drawer_ours_vlm_10_demos_v2_11_acting.sh
./train_put_item_in_drawer_ours_vlm_10_demos_v2_11_stabilizing.sh

# validation/test
./eval_put_item_in_drawer_ours_vlm_10_demos_v2_11.sh

Hand Over Item

# training
./train_hand_over_item_ours_vlm_10_demos_v1_11_acting.sh
./train_hand_over_item_ours_vlm_10_demos_v1_11_stabilizing.sh

# validation/test
./eval_hand_over_item_ours_vlm_10_demos_v1_11.sh

How to generate datasets

First, make sure you're inside the voxactb/RLBench/tools folder.

Open Jar

# training data
python dataset_generator_two_robots.py --tasks=open_jar \
                            --save_path=$PERACT_ROOT/data/train \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=10 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=open_jar_noises_starting_states_dominant_assistive

# val data
python dataset_generator_two_robots.py --tasks=open_jar \
                            --save_path=$PERACT_ROOT/data/val \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=open_jar_close_to_jar_dominant_assistive \
                            --seed=43

# test data
python dataset_generator_two_robots.py --tasks=open_jar \
                            --save_path=$PERACT_ROOT/data/test \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=open_jar_close_to_jar_dominant_assistive \
                            --seed=88

Open Drawer

# training data
python dataset_generator_two_robots.py --tasks=open_drawer \
                            --save_path=$PERACT_ROOT/data/train \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=10 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=open_drawer_noises_starting_states_dominant_assistive

# val data
python dataset_generator_two_robots.py --tasks=open_drawer \
                            --save_path=$PERACT_ROOT/data/val \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=open_drawer_close_to_drawer_dominant_assistive \
                            --seed=43

# test data
python dataset_generator_two_robots.py --tasks=open_drawer \
                            --save_path=$PERACT_ROOT/data/test \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=open_drawer_close_to_drawer_dominant_assistive \
                            --seed=88

Put Item in Drawer

# training data
python dataset_generator_two_robots.py --tasks=put_item_in_drawer \
                            --save_path=$PERACT_ROOT/data/train \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=10 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=put_item_in_drawer_noises_starting_states_dominant_assistive											
# val data
python dataset_generator_two_robots.py --tasks=put_item_in_drawer \
                            --save_path=$PERACT_ROOT/data/val \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=put_item_in_drawer_noises_starting_states_dominant_assistive \
                            --seed=43

# test data
python dataset_generator_two_robots.py --tasks=put_item_in_drawer \
                            --save_path=$PERACT_ROOT/data/test \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=put_item_in_drawer_noises_starting_states_dominant_assistive \
                            --seed=88

Hand Over Item

# training data
python dataset_generator_two_robots.py --tasks=hand_over_item \
                            --save_path=$PERACT_ROOT/data/train \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=10 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=hand_over_item_noises_starting_states_dominant_assistive

# val data
python dataset_generator_two_robots.py --tasks=hand_over_item \
                            --save_path=$PERACT_ROOT/data/val \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=hand_over_item_noises_starting_states_dominant_assistive \
                            --seed=43

# test data
python dataset_generator_two_robots.py --tasks=hand_over_item \
                            --save_path=$PERACT_ROOT/data/test \
                            --image_size=128,128 \
                            --renderer=opengl \
                            --episodes_per_task=25 \
                            --processes=1 \
                            --all_variations=True \
                            --mode=hand_over_item_noises_starting_states_dominant_assistive \
                            --seed=88

Citations

VoxAct-B

@inproceedings{liu2024voxactb,
    title={VoxAct‐B: Voxel‐Based Acting and Stabilizing Policy for Bimanual Manipulation},
    author={I-Chun Arthur Liu and Sicheng He and Daniel Seita and Gaurav S. Sukhatme},
    booktitle={Conference on Robot Learning},
    year={2024}
}

PerAct

@inproceedings{shridhar2022peract,
    title     = {Perceiver-Actor: A Multi-Task Transformer for Robotic Manipulation},
    author    = {Shridhar, Mohit and Manuelli, Lucas and Fox, Dieter},
    booktitle = {Proceedings of the 6th Conference on Robot Learning (CoRL)},
    year      = {2022},
}

C2FARM

@inproceedings{james2022coarse,
    title={Coarse-to-fine q-attention: Efficient learning for visual robotic manipulation via discretisation},
    author={James, Stephen and Wada, Kentaro and Laidlow, Tristan and Davison, Andrew J},
    booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
    pages={13739--13748},
    year={2022}
}

PerceiverIO

@article{jaegle2021perceiver,
    title={Perceiver io: A general architecture for structured inputs \& outputs},
    author={Jaegle, Andrew and Borgeaud, Sebastian and Alayrac, Jean-Baptiste and Doersch, Carl and Ionescu, Catalin and Ding, David and Koppula, Skanda and Zoran, Daniel and Brock, Andrew and Shelhamer, Evan and others},
    journal={arXiv preprint arXiv:2107.14795},
    year={2021}
}

RLBench

@article{james2020rlbench,
    title={Rlbench: The robot learning benchmark \& learning environment},
    author={James, Stephen and Ma, Zicong and Arrojo, David Rovick and Davison, Andrew J},
    journal={IEEE Robotics and Automation Letters},
    volume={5},
    number={2},
    pages={3019--3026},
    year={2020},
    publisher={IEEE}
}

VoxPoser

@article{huang2023voxposer,
    title={VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models},
    author={Huang, Wenlong and Wang, Chen and Zhang, Ruohan and Li, Yunzhu and Wu, Jiajun and Fei-Fei, Li},
    journal={arXiv preprint arXiv:2307.05973},
    year={2023}
}