Explainable and Sensor-Adaptable Recurrent Networks (ExSARN)

This repository contains tactile data and material description for texture classification, supplementary to the paper "On Explainability and Sensor-Adaptability of a Robot Tactile TextureRepresentation Using a Two-Stage Recurrent Networks", R. Gao et al., accepted to IROS 2021

For easy implementation, we only include the compiled dataset here in folder data/. The data collection process follows the setting in Supervised Autoencoder Joint Learning on Heterogeneous Tactile Sensory Data and more details about the robot arm and sensor specs can be found there.

Dataset description

This dataset contains 50 classes. A snapshot of each corresponding material is shown below and the detailed description for each material can be found in docs/material_properties_c50.docx text file. Materials are named by Matx, where x is an index number ranging from 1 to 50. They are arranged from left to right, top to bottom in snapshots. Note that labels in datasets are reduced by 1 so that labels start from index 0 for programming purpose.

We store the data in Google Drive. You can either download the data directly from Google Drive or use the provided scripts to fetch it conveniently from terminal. Once cloning this repository, please do the following:

1. Modify the dir of "data_temp" to the desired directory
2. Set fetch_data.sh runnable by chmod +x fetch_data.sh
3. Set Python script gdrivedl.py as runnable program
4. In terminal, run ./fetch_data.sh to download tactile data from Google Drive

The data set is ready for quick implementation in PyTorch. The data can be loaded with following commands

import torch
from preprocess_data import*

args_data_dir = 'data/'
data_folder = "c20/"
dataset_dir = os.path.join(args_data_dir, data_folder)
batch_size = 32
shuffle = True
prefix = ""
sequence_length = 75 if data_folder=="c20icub/" else 400
data_type = "I" if data_folder=="c20icub/" else "B"
train_loader, val_loader, test_loader = get_TrainValTestLoader(dataset_dir, k=0, batch_size=batch_size,shuffle=shuffle, prefix=prefix,seq_len=sequence_length, data_type=data_type)

'''epoch and model setting here'''

for i, (x,  y) in enumerate(train_loader):
    '''model training here'''

Distribution

If you find our data / method helpful, please cite our paper. Suggestions, discussion, and collaboration are always welcomed!

@inproceedings{gao2021explainability,
title = {On Explainability and Sensor-Adaptability of a Robot Tactile Texture Representation Using a Two-Stage Recurrent Networks},
author = {Ruihan Gao and Tian Tian and Zhiping Lin and Yan Wu},
year = {2021},
date = {2021-09-27},
booktitle = {2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) },
keywords = {},
pubstate = {forthcoming},
tppubtype = {inproceedings}
}