FlexWear-HD_Dataset/README.md

FlexWear-HD

Abstract

This dataset includes high-density electromyography (HDEMG) data from 13 users without motor disabilities for 10 common gestures. Two sessions per user is available, with 8-10 gesture trials per gesture performed in the first session and 4-5 gestures trials per gesture performed in the second session. About one hour passes between sessions, and the sensor is kept on between the two sessions. The sensor used is an easy-to-wear reusable forearm device that uses 64 hydrogel electrodes. There are 960,000-1,200,000 time steps provided per subject while sampling with a sampling rate of 4000 Hz. Data is saved in the compressed HDF5 format, with two files per subject (one file per session).

This data can be used to train EMG-based gesture classifiers for control of computers or robots. Additional information on the sensor and data collection is available at https://arxiv.org/abs/2312.07745, which is a paper that also uses this data for control of an 8 degree-of-freedom mobile manipulator.

File Format and Variables

The ten categories of gestures are labeled with the following keys in the HDF5 file:

1. abduct_p1 (wrist abduction),
2. adduct_p1 (wrist adduction),
3. extend_p1 (finger abduction and extension),
4. grip_p1 (fist),
5. pronate_p1 (wrist pronation),
6. rest_p1 (rest),
7. supinate_p1 (wrist supination),
8. tripod_p1 (thumb, index, and middle finger pinch),
9. wextend_p1 (wrist extension),
10. wflex_p1 (wrist flexion). These variables contain the EMG data. These variables include data in a 3D array format of dimensions (trial, electrode, timestep).

EMG data from each session is saved as a different HDF5 file. Each user's data is saved in a separate folder, with participant folders labeled from p001 to p013. The first session has the suffix initial and the second session has the suffix recalibration.

Additional keys include SNR and Impedance_p0. SNR includes a single float64 that is calculated from the root-mean-squared (RMS) of maximum voluntary contraction during a fist gesture divided by the RMS of the rest gesture. Impedance_p0 includes 64 float64 numbers based on the measured impedance from the separate electrodes to ground.