EEGdash›NeMAR›NM000106

Iss. 106 · 100 subjects · 807 recordings · CC-BY-NC 4.0

Dataset Brief · FRL Handwriting

NM000106: emg dataset, 100 subjects#

FRL Handwriting: Handwriting Decoding from Surface Electromyography

Access recordings and metadata through EEGDash.

Citation: Patrick Kaifosh, Thomas R. Reardon, CTRL-labs at Reality Labs (2025). FRL Handwriting: Handwriting Decoding from Surface Electromyography. 10.82901/nemar.nm000106

Modality: emg Subjects: 100 Recordings: 807 License: CC-BY-NC 4.0 Source: nemar

Metadata: Complete (100%)

100-participant EMG dataset — FRL Handwriting: Handwriting Decoding from Surface Electromyography.

Data & curation Patrick Kaifosh · Thomas R. Reardon · CTRL-labs at Reality Labs
Year 2025 · Distributed via NeMAR
Funding Meta Reality Labs

EMG · 16 ch2000 HzBIDS 1.11.0Task · handwriting26 sessions

Layer 01Study

What was asked

Hypothesis, independent & dependent variables, paradigm, cohort, and the editorial caveats around what the recordings can and cannot answer.

Layer 02Signal · BIDS

What was recorded

Sidecars, channels & electrodes, coordinate system, event semantics, and quality stats from the NEMAR pipeline when available.

Layer 03Training · ML

What you can train on

Recommended access modes — MNE Raw, braindecode windows, PyTorch DataLoader — plus the targets the metadata makes addressable.

§ 01Access · Get started

Quickstart#

Get Started

Install

pip install eegdash

Access the data

from eegdash.dataset import NM000106

dataset = NM000106(cache_dir="./data")
# Get the raw object of the first recording
raw = dataset.datasets[0].raw
print(raw.info)

Query & Filter

Filter by subject

dataset = NM000106(cache_dir="./data", subject="01")

Advanced query

dataset = NM000106(
    cache_dir="./data",
    query={"subject": {"$in": ["01", "02"]}},
)

Iterate recordings

for rec in dataset:
    print(rec.subject, rec.raw.info['sfreq'])

Cite This Dataset

If you use this dataset in your research, please cite the original authors.

BibTeX

@dataset{nm000106,
  title = {FRL Handwriting: Handwriting Decoding from Surface Electromyography},
  author = {Patrick Kaifosh and Thomas R. Reardon and CTRL-labs at Reality Labs},
  doi = {10.82901/nemar.nm000106},
  url = {https://doi.org/10.82901/nemar.nm000106},
}

§ 02Study · The README

About This Dataset#

Dataset: handwriting - Imagined handwriting from wrist-based surface electromyography

Task: Air-writing (imagined handwriting without pen) Participants: 100 subjects Sessions: ~700 total (~7 per subject) Publication: Kaifosh et al., 2025 - “A generic non-invasive neuromotor interface for human-computer interaction” (Nature)

This dataset captures wrist-based sEMG signals during imagined handwriting motions for text entry. Participants “write” prompted text with fingers together (as if holding an invisible pen) without any physical writing surface. Applications include AR/VR text input, mobile computing, and hands-free communication.

handwriting: Handwriting Recognition from EMG

Overview

Dataset Details

Participants

View full README

handwriting: Handwriting Recognition from EMG

Overview

Dataset Details

Participants

Sample size: 100 participants
Demographics: Not available (marked as n/a)
Recording side: Dominant wrist
Sessions: Average 7 per participant

Hardware

Device: sEMG-RD (single wristband)
Channels: 16 (EMG0-EMG15)
Sampling rate: 2000 Hz
Reference: Bipolar differential

Recording Protocol

Participant holds fingers together (as if holding pen)
Prompted text appears on screen
Participant “writes” the text in air
Session duration: ~11 minutes
Prompts per session: 96 phrases

Data Contents

Files per Session

sub-XXX/ses-XXX/emg/

├── sub-XXX_ses-XXX_task-handwriting_emg.edf
├── sub-XXX_ses-XXX_task-handwriting_emg.json
├── sub-XXX_ses-XXX_task-handwriting_channels.tsv
├── sub-XXX_ses-XXX_task-handwriting_events.tsv
└── sub-XXX_ses-XXX_electrodes.tsv

Events

Handwriting prompts: Text to be written - prompt_text: Displayed phrase
Stage boundaries: Posture changes (sitting/standing), session phases

Coordinate System

Single coordinate system at root (dominant wrist, percent units, no decimals)

Baseline Performance

Published Results (Kaifosh et al., 2025)

Generic Model (6,527 training participants): - Offline CER: >90% classification accuracy on held-out participants - Online performance: 20.9 words per minute (WPM) - Online CER: Median improvement from ~35% (practice) to ~25% (evaluation)

Personalized Model (20 min fine-tuning): - 16% improvement over generic model - Better performance for users with higher generic CER - Diminishing returns with more pretraining data

Comparison: - Open-loop handwriting (no pen): 25.1 WPM - sEMG handwriting: 20.9 WPM (83% of baseline) - Mobile phone keyboard: 36 WPM

Model architecture: MPF features + Conformer (attention mechanism)

Use Cases

Keyboard-free text entry: AR/VR, mobile devices
Silent communication: Private text input in public spaces
Personalization research: Few-shot learning, transfer learning
Sequence modeling: Character-level prediction with attention

Known Limitations

Single wrist (dominant hand only)
Handedness not recorded
Learning curve: Users improve with practice/coaching
Lower WPM than physical writing or typing

Citation

Kaifosh, P., Reardon, T.R., & CTRL-labs at Reality Labs. (2025).
A generic non-invasive neuromotor interface for human-computer interaction.
Nature, 645(8081), 702-711. https://doi.org/10.1038/s41586-025-09255-w

Data Curator

Yahya Shirazi SCCN (Swartz Center for Computational Neuroscience) INC (Institute for Neural Computation) University of California San Diego

Version History

v1.0 (2025-10-01): Initial BIDS conversion

BIDS Version: 1.11 | EMG-BIDS: BEP-042 | Updated: Oct 1, 2025

NEMAR Metadata#

# handwriting: Handwriting Recognition from EMG ## Overview Dataset: handwriting - Imagined handwriting from wrist-based surface electromyography Task: Air-writing (imagined handwriting without pen) Participants: 100 subjects Sessions: ~700 total (~7 per subject) Publication: Kaifosh et al., 2025 - “A generic non-invasive neuromotor interface for human-computer interaction” (Nature) ### Purpose This dataset captures wrist-based sEMG signals during imagined handwriting motions for text entry. Participants “write” prompted text with fingers together (as if holding an invisible pen) without any physical writing surface. Applications include AR/VR text input, mobile computing, and hands-free communication. ## Dataset Details ### Participants - Sample size: 100 participants - Demographics: Not available (marked as n/a) - Recording side: Dominant wrist - Sessions: Average 7 per participant ### Hardware - Device: sEMG-RD (single wristband) - Channels: 16 (EMG0-EMG15) - Sampling rate: 2000 Hz - Reference: Bipolar differential ### Recording Protocol 1. Participant holds fingers together (as if holding pen) 2. Prompted text appears on screen 3. Participant “writes” the text in air 4. Session duration: ~11 minutes 5. Prompts per session: 96 phrases ## Data Contents ### Files per Session ` sub-XXX/ses-XXX/emg/ ├── sub-XXX_ses-XXX_task-handwriting_emg.edf ├── sub-XXX_ses-XXX_task-handwriting_emg.json ├── sub-XXX_ses-XXX_task-handwriting_channels.tsv ├── sub-XXX_ses-XXX_task-handwriting_events.tsv └── sub-XXX_ses-XXX_electrodes.tsv ` ### Events - Handwriting prompts: Text to be written

prompt_text: Displayed phrase

Stage boundaries: Posture changes (sitting/standing), session phases

### Coordinate System Single coordinate system at root (dominant wrist, percent units, no decimals) ## Baseline Performance ### Published Results (Kaifosh et al., 2025) Generic Model (6,527 training participants): - Offline CER: >90% classification accuracy on held-out participants - Online performance: 20.9 words per minute (WPM) - Online CER: Median improvement from ~35% (practice) to ~25% (evaluation) Personalized Model (20 min fine-tuning): - 16% improvement over generic model - Better performance for users with higher generic CER - Diminishing returns with more pretraining data Comparison: - Open-loop handwriting (no pen): 25.1 WPM - sEMG handwriting: 20.9 WPM (83% of baseline) - Mobile phone keyboard: 36 WPM Model architecture: MPF features + Conformer (attention mechanism) ## Use Cases - Keyboard-free text entry: AR/VR, mobile devices - Silent communication: Private text input in public spaces - Personalization research: Few-shot learning, transfer learning - Sequence modeling: Character-level prediction with attention ## Known Limitations - Single wrist (dominant hand only) - Handedness not recorded - Learning curve: Users improve with practice/coaching - Lower WPM than physical writing or typing ## Citation ` Kaifosh, P., Reardon, T.R., & CTRL-labs at Reality Labs. (2025). A generic non-invasive neuromotor interface for human-computer interaction. Nature, 645(8081), 702-711. https://doi.org/10.1038/s41586-025-09255-w ` ## Data Curator Yahya Shirazi SCCN (Swartz Center for Computational Neuroscience) INC (Institute for Neural Computation) University of California San Diego ## Version History v1.0 (2025-10-01): Initial BIDS conversion — BIDS Version: 1.11 | EMG-BIDS: BEP-042 | Updated: Oct 1, 2025

License: CC-BY-NC 4.0

Authors:

Patrick Kaifosh
Thomas R. Reardon
CTRL-labs at Reality Labs

Versions:

Version	DOI	Released
`current`	10.82901/nemar.nm000106

§ 03Cohort · Participants

Cohort#

Dataset Statistics#

Channel counts: 16 ch (n=807 recordings)

Sampling frequencies: 2000.0 Hz (n=807 recordings)

Total recording duration: 140 h

§ 04Signal · Electrodes & trace

Signal · Electrodes & live trace#

Fig. 01 Signal & montage 16 ch · EMG · 2000 Hz · 100 subjects, 807 recordings

Live trace viewer — sub-021 · ses-008 · task-handwriting

Showing one representative recording out of 100 subjects and 807 recordings in this dataset. Browse the full set on OpenNeuro; drop any other _emg.{set,edf,bdf,vhdr} file onto the viewer (or pass ?emg=<url>) to inspect it.

Electrode layout — EMG · 16 sensors — 16 channels

NEMAR Processing Statistics#

The plots below are generated by NEMAR’s automated EEG pipeline. The histogram shows pipeline success for data cleaning and ICA decomposition, the percentage of data frames and EEG channels retained after artefact removal, line noise per channel (RMS, dB), and the age/gender distribution of participants.

HED event descriptors word cloud

§ 05Manifest · BIDS tree

Manifest#

File Explorer#

Browse the BIDS file structure of this dataset. Records are fetched on demand from the EEGDash catalog the first time you open the explorer.

Recordings—

Files—

Subjects—

Modalities—

Click to load file structure…

§ 06API · Programmatic access

API Reference#

Signature

eegdash.dataset

class

eegdash.dataset.NM000106(cache_dir, query=None, s3_bucket=None, **kwargs)

Bases: EEGDashDataset

Author (year)Kaifosh2025_106

Canonical—

Importable asNM000106 · Kaifosh2025_106

Sourceeegdash/dataset/registry.py · [source ↗]

class eegdash.dataset.NM000106(cache_dir: str, query: dict | None = None, s3_bucket: str | None = None, **kwargs)[source]#

FRL Handwriting: Handwriting Decoding from Surface Electromyography

Study:: nm000106 (NeMAR)
Author (year):: Kaifosh2025_106
Canonical:: —

Also importable as: NM000106, Kaifosh2025_106.

Modality: emg; Subject type: Unknown. Subjects: 100; recordings: 807; tasks: 1.

Parameters:

cache_dir (str | Path) – Directory where data are cached locally.
query (dict | None) – Additional MongoDB-style filters to AND with the dataset selection. Must not contain the key dataset.
s3_bucket (str | None) – Base S3 bucket used to locate the data.
**kwargs (dict) – Additional keyword arguments forwarded to EEGDashDataset.

data_dir#

Local dataset cache directory (cache_dir / dataset_id).

Type:: Path

query#

Merged query with the dataset filter applied.

Type:: dict

records#

Metadata records used to build the dataset, if pre-fetched.

Type:: list[dict] | None

Notes

Each item is a recording; recording-level metadata are available via dataset.description. query supports MongoDB-style filters on fields in ALLOWED_QUERY_FIELDS and is combined with the dataset filter. Dataset-specific caveats are not provided in the summary metadata.

References

OpenNeuro dataset: https://openneuro.org/datasets/nm000106 NeMAR dataset: https://nemar.org/dataexplorer/detail?dataset_id=nm000106 DOI: https://doi.org/10.82901/nemar.nm000106

Examples

>>> from eegdash.dataset import NM000106
>>> dataset = NM000106(cache_dir="./data")
>>> recording = dataset[0]
>>> raw = recording.load()

__init__(cache_dir: str, query: dict | None = None, s3_bucket: str | None = None, **kwargs)[source]#

save(path: str, overwrite: bool = False, offset: int = 0)[source]#

Save datasets to files by creating one subdirectory for each dataset:

path/
    0/
        0-raw.fif | 0-epo.fif
        description.json
        raw_preproc_kwargs.json (if raws were preprocessed)
        window_kwargs.json (if this is a windowed dataset)
        window_preproc_kwargs.json  (if windows were preprocessed)
        target_name.json (if target_name is not None and dataset is raw)
    1/
        1-raw.fif | 1-epo.fif
        description.json
        raw_preproc_kwargs.json (if raws were preprocessed)
        window_kwargs.json (if this is a windowed dataset)
        window_preproc_kwargs.json  (if windows were preprocessed)
        target_name.json (if target_name is not None and dataset is raw)

Parameters:

path (str) –

Directory in which subdirectories are created to store
-raw.fif | -epo.fif and .json files to.
overwrite (bool) – Whether to delete old subdirectories that will be saved to in this call.
offset (int) – If provided, the integer is added to the id of the dataset in the concat. This is useful in the setting of very large datasets, where one dataset has to be processed and saved at a time to account for its original position.

Access modesMNE → braindecode → PyTorch → ML

.rawMNE Raw object — standard tools (filter, epoch, ICA, plot_psd).mne

BaseConcatDatasetEach record is a lazy BaseDataset from braindecode — windowed via create_windows_from_events.braindecode

DataLoaderWraps the windowed dataset into a PyTorch DataLoader; supports parallel workers and on-the-fly augmentations.pytorch

Zarr cacheOptional braindecode Zarr mirror for fast resume; persisted to cache_dir.zarr

Hugging FacePre-bundled mirror at EEGDash/nm000106 · pull with datasets.load_dataset("EEGDash/nm000106").huggingface

Croissant 1.0Machine-readable JSON-LD descriptor — NM000106.croissant.json (MLCommons schema, ingestible by PyTorch / TensorFlow / JAX).mlcommons

Examples using EEGDashcurated · start here

Find datasets with the EEGDash APIQuery the catalogue, filter by task or modality, list candidates.

Load one EEG recordingResolve a single record to an MNE Raw with channels and events.

EEG recording to PyTorch DataLoaderWrap braindecode windows in a DataLoader for model training.

Preprocess EEG and create windowsFilter, resample, epoch — and persist the windowed dataset.

Save and reload prepared dataCache a windowed dataset to disk and reattach it without recompute.

Download a dataset locallyPrefetch BIDS files to a local cache and validate the layout.

Swap any load_dataset(...) call for nm000106 to reproduce the tutorial on this dataset.

Citation

Patrick Kaifosh, Thomas R. Reardon, CTRL-labs at Reality Labs (2025). FRL Handwriting: Handwriting Decoding from Surface Electromyography. 10.82901/nemar.nm000106

Provenance

¹Contributed to nemar in BIDS format.

²Curated & ingested by the EEGDash catalog; see CITATION.cff for canonical reference.

³Persistent identifier: 10.82901/nemar.nm000106.

Related & sibling datasets

NM000104EMG · 108 subj NM000107EMG · 100 subj NM000105EMG · 100 subj NM000159EMG · 16 subj NM000155EMG · 6 subj

+ 1 more — see See Also below →

BIDS

BIDS 1.11.0

Sidecars

events · channels · electrodes

Provenance

CC-BY-NC 4.0 · 10.82901/nemar.nm000106

Machine-readable

schema.org/Dataset · Croissant

Mirrors

OpenNeuro · NEMAR · HuggingFace · Paper

Dataset ID	`NM000106`
Title	FRL Handwriting: Handwriting Decoding from Surface Electromyography
Author (year)	`Kaifosh2025_106`
Canonical	—
Importable as	`NM000106`, `Kaifosh2025_106`
Year	2025
Authors	Patrick Kaifosh, Thomas R. Reardon, CTRL-labs at Reality Labs
License	CC-BY-NC 4.0
Citation / DOI	10.82901/nemar.nm000106
Source links	OpenNeuro \| NeMAR

NM000106: emg dataset, 100 subjects#

Quickstart#

About This Dataset#

NEMAR Metadata#

Cohort#

Dataset Statistics#

Signal · Electrodes & live trace#

NEMAR Processing Statistics#

Manifest#

File Explorer#

API Reference#

Citation

Provenance

Related & sibling datasets

See Also#