KROCO is an acronym for Kinesthetic Robot Contacts. The dataset consists of Learning from Demonstration (LfD, aka Programming by Demonstration (PbD)) samples that were collected via kinesthetic teaching. Each labeled sample is a contact-based (force-based) interaction with the environment. The dataset is in hdf5 format.
- Download Link
- Digital Object Identifier (DOI): nhea-gt59
Dataset Description
Tools
Environments
Structure
The hierarchical dataset is structured as follows:
.
+ data # a group containing all datasets
+ 0 # a dataset for one interaction
+ attrs: label, label_string, tool, environment, hash,
+ 1
+ ...
+ labels # a label vector for all classes
Labels
Class ID | Name |
---|---|
0 | touch |
1 | press |
2 | press and slide |
3 | contour |
4 | insert |
5 | hand-over |
6 | turn |
7 | push |
Dimensions
Each dataset is in the form of (N x D) with N samples and D=15 dimensions. N differs between datasets depending on the duration of the interaction. The following table shows the 0-indexed data columns and the unit in brackets [].
Data column | Description |
---|---|
0..2 | position x,y,z in [m] |
3..6 | orientation quaternions in scalar-last (x,y,z,w) format |
7..12 | wrench in origin frame (f_x, f_y, f_z, m_x, m_y, m_z) [N, Nm] |
13 | gripper finger distance from 0..1, 0: fully open, 1: fully closed |
14 | grasp status (-1: no object in gripper, 0: gripper moving, 1: object in gripper) |
Sample Rate
The sample rate is 50 Hz.
Read Dataset
Python
There is code in the repository (python) to load the hdf5
dataset. Adapt the variable file_path
to the destination of the dataset.
Matlab
An example to display the whole dataset and read the first segment:
h5disp('DATASET_PATH', '/')
dataset0 = h5read('DATASET_PATH', '/data/0')
For more details, please check this guide.
Reference
If you find this dataset useful for your research, please consider citing it with the given reference
@InProceedings{eiband2019identification,
author = {Eiband, Thomas and Lee, Dongeui},
booktitle = {IEEE-RAS 2020 International Conference on Humanoid Robots (Humanoids)},
date = {2021},
title = {Identification of Common Force-based Robot Skills from the Human and Robot Perspective},
organization = {IEEE},
}