PanoVine: Whole-Body Visuomotor Control for Soft Growing Vine Robot

[Project page] [Paper] [Video]

Yimeng Qin*, Xiaomeng Xu*, William Heap, Aditi Oak, Shuran Song†, Allison Okamura†

*Equal Contribution, †Equal Advising

Stanford University

Clone this repo:

git clone https://github.com/xxm19/panovine.git
cd panovine

Create the mamba environment (We recommend Mambaforge over the standard Anaconda distribution for a faster environment solve):

mamba env create -f conda_environment.yaml
conda activate panovine

On-robot data collection and inference (unified_data_logger.py, policy_inference.py) additionally require a ROS Noetic installation providing rospy/std_msgs, plus the vine robot motor and sensor drivers. These are not needed for offline data processing or policy training, which run from the conda environment alone.

python unified_data_logger.py

saved file structure:

• robot_logs/baseStation/<id>/*_sensors.npz (base encoder, pressure, torque)
• robot_logs/controller2/<id>/*_sensors.npz (imu_id, encoder1/2, roll/pitch/yaw)
• robot_logs/camera/<id>/cam_*.{mp4,avi} + timestamp csv/json

python process_data_parallel.py \
  --log-root robot_logs \
  --output-path dataset.zarr.zip \
  --num-workers 16 \
  --action-repr angle

The training config is:

diffusion_policy/config/train_diffusion_transformer_free_space_vine_workspace.yaml

Run training:

python train.py \
  --config-name train_diffusion_transformer_free_space_vine_workspace \
  task.dataset_path=dataset.zarr.zip

multi-gpu

accelerate launch --num_processes 4 \
  train.py --config-name train_diffusion_transformer_free_space_vine_workspace \
  task.dataset_path=dataset.zarr.zip

python policy_inference.py \
  --config-name train_diffusion_transformer_free_space_vine_workspace \
  input=/path/to/ckpt.ckpt

@article{qin2026panovine,
  title={PanoVine: Whole-Body Visuomotor Control for Soft Growing Vine Robot},
  author={Qin, Yimeng and Xu, Xiaomeng and Heap, William and Oak, Aditi and Song, Shuran and Okamura, Allison},
  journal={arXiv preprint arXiv:2606.22923},
  year={2026}
}

This repository is released under the MIT license. See LICENSE for more details.

• Our policy implementation is adapted from RoboPanoptes and UMI.