license: mit
tags:
- BioTac
- DIGIT
- Tactile sensors
- Transfer learning
pretty_name: ACROSS Dataset
size_categories:
- 100B<n<1T
ACROSS: A Deformation-Based Cross-Modal Representation for Robotic Tactile Perception
Accepted to 2025 IEEE Conference on Robotics and Automation (ICRA 2025)
Paper page can be found here.
Github Repository can be found here.
DIGIT BioTac Isaac Gym.
This package contains the dataset for the BioTac to Digit pipeline. It includes over 155K unique 3D mesh deformation pairs from interactions involving BioTac and DIGIT sensors. The dataset covers various types of indenters and the forces applied to each sensor.
This dataset is simulated with Isaac Gym. To resimulate the dataset or extand it, feel you can use the code provided on our github repository.
Dataset Structure
The dataset is structured as follows:
results_biotac_[index].hdf5
results_digit_[index].hdf5
The indexes are used to match the pairs of BioTac and DIGIT sensors. The dataset is split into 3 parts.
Each file contains 100 unique trajectories with 9 different indenters.
Each trajectory contains n steps and the 6D transformation of the sensor.
Each step contains the following information:
'max_contact_distances', 'max_contact_distances_pos', 'net_force_vecs', 'nodal_coords'
Citation
@InProceedings{ZaiElAmri2025ACROSS,
author = {Zai El Amri, Wadhah and Kuhlmann, Malte and {Navarro-Guerrero}, Nicol{\'a}s},
title = {{{ACROSS}}: {{A Deformation-Based Cross-Modal Representation}} for {{Robotic Tactile Perception}}},
booktitle = {{{IEEE International Conference}} on {{Robotics}} and {{Automation}} ({{ICRA}})},
year={2025},
}