Tactile Data of 12 Textures on Uneven Surfaces Collected with a 6-DoF Robotic Arm

Published: 18 August 2026| Version 1 | DOI: 10.17632/yctpcddnm7.1
Contributors:
,
,
,
,

Description

This dataset provides tactile data captured with the BioIn-Tacto multimodal tactile sensing module [1, 2] mounted on the end-effector of a Lite6 robotic arm [3]. It includes barometric and MARG (Magnetic, Angular Rate, and Gravity) data to support research in texture recognition and robotic manipulation, collected as the manipulator moved the sensing module across 12 textures applied to concave/convex surfaces. The data is organized into folders representing the processing stages: Data/ ├── 0_Raw: Raw CSVs extracted from ROS bags, organized by shape/texture(T1–T12)/experiment (25 movements per texture). ├── 1_Merged: Raw per-sensor CSVs merged into a single time-aligned CSV per experiment. ├── 2_Trimmed: Recordings trimmed to keep only the valid interaction segment. ├── 3_Normalized: Trimmed signals normalized (scaled) per experiment, producing normalized_data.csv files. ├── 4_Windowed: Signals split into sliding windows of 128, 256, and 512 samples. ├── 5_Reindexed: Windows reindexed and merged for consistent numbering across experiments. ├── 6_NPY: Final .npy arrays per dataset and window size, plus windows_per_experiment.json metadata. └── 7_Folded: Windows organized into cross-validation folds for train/test evaluation. Each exploratory movement contains baro.csv (barometric data) and imus.csv (IMU data: acceleration, angular rate, and magnetic field). The `Scripts` folder includes tools for automating data preprocessing: Scripts/ ├── merge_datasets.sh: Runs merge.py per shape to build 1_Merged. ├── trim_dataset.sh: Trims recordings; also supports --create_json_only / --use_json modes and per-texture or preview runs. ├── normalize_datasets.sh: Runs normalize.py on every trimmed CSV to build 3_Normalized. ├── create_windows.sh: Runs window_creator.py per dataset/surface to generate the 128/256/512-sample windows. ├── reindex_windows.sh: Runs reindex_windows.py to reindex and merge windows into 6_Reindexed. ├── create_npy.sh: Runs npy_creator.py per dataset and window size to export .npy files. ├── generate_windows_metadata.sh: Runs generate_windows_metadata.py to write per-dataset windows_per_experiment.json files. ├── fold_generator.py: creates the balanced 2, 4, and 5-Fold cross-validation splits from the windowed .npy data. └── run.sh: Automates the entire data preprocessing pipeline. It sequentially runs the previous steps. [1] T. E. Alves de Oliveira, A. -M. Cretu and E. M. Petriu, "Multimodal Bio-Inspired Tactile Sensing Module," in IEEE Sensors Journal, vol. 17, no. 11, pp. 3231-3243, 1 June1, 2017, https://doi.org/10.1109/JSEN.2017.2690898. [2] T. E. Alves de Oliveira, V. Prado da Fonseca, BioIn-Tacto: A compliant multi-modal tactile sensing module for robotic tasks, HardwareX, Volume 16, 2023, e00478, ISSN 2468-0672, https://doi.org/10.1016/j.ohx.2023.e00478. [3] Ltd. Shenzhen UFACTORY Co., UFACTORY Lite 6 User Manual, (n.d.). https://static.generation-robots.com/media/ufactory-lite6-user-manual.pdf (accessed August 10, 2026).

Files

Steps to reproduce

Data collection was performed with a 6-DoF Lite6 manipulator carrying the BioIn-Tacto sensing module on its end-effector, sliding it across textures bonded to curved surfaces. The exploration trajectory for each surface is generated from an initial probing phase, the robot making contact with the surface at twelve reference points, recording the IMU orientation at the instant the barometer detects contact pressure, so that the measured inclination closely approximates the local surface normal. These normals drive the trajectory generation, yielding sliding paths that follow the surface geometry with consistent contact throughout. The textures explored in this dataset consisted of brocade fabric, carpet wool, embossed plastic, honeycomb fabric, mesh cotton, mesh leather, open weave cotton, reptile-patterned leather, ridged polymer, silicone mesh, tight weave cotton, and wood.

Institutions

Categories

Robotics, Microstructural Texture, Tactile Perception, Surface Texture, Time Series, Measurement in Robotics

Funders

Licence