ProSoRo-100K

This dataset contains 100K samples of data for six different soft robotic structures: cylinder, dome, neck, octagonal prism, origami, and quadrangular prism. They were generated by finite element simulations, and used for training ProSoRo-MVAE.

Dataset Schema

There are six subsets in this dataset, one for each soft robotic structure. Each subset contains the following features:

Field Name	Type	Shape	Description
motion	List[float64]	[6]	The 6D motion of the ProSoRo, including translation (dx, dy, dz) and rotation (rx, ry, rz) in mm and rad.
force	List[float64]	[6]	The 6D force and torque on the bottom surface of the ProSoRo, corresponding to (fx, fy, fz, tx, ty, tz) in N and Nmm.
nodes	List[List[float64]]	[N,3]	The 3D displacement of N surface nodes of the ProSoRo, where each node is represented as [dx, dy, dz] in mm.

Usage

from datasets import load_dataset

dataset = load_dataset("asRobotics/prosoro-100k")

# Access the cylinder subset
for sample in dataset['cylinder']:
    print(sample['motion'], sample['force'], sample['nodes'])

Citation

If you use this model in your research, please cite the following paper:

@article{han2025anchoring,
    title={Anchoring Morphological Representations Unlocks Latent Proprioception in Soft Robots},
    author={Han, Xudong and Guo, Ning and Xu, Ronghan and Wan, Fang and Song, Chaoyang},
    journal={Advanced Intelligent Systems},
    volume={0},
    pages={0-0},
    year={2025}
}