{"dataset":{"id":"50","dataset_id":"nm000105","name":"FRL Discrete Gestures: Hand Gesture Recognition from Surface Electromyography","description":"NEMAR Dataset nm000105: discrete_gestures - Discrete hand gesture detection from EMG","owner_user_id":2,"status":"active","github_repo":"nemarDatasets/nm000105","concept_doi":"10.82901/nemar.nm000105","latest_version_doi":"10.82901/nemar.nm000105.v2.0.0","created_at":"2026-01-19 03:25:23","updated_at":"2026-07-10 21:42:53","zenodo_concept_id":null,"is_sandbox":0,"visibility":"public","ezid_status":"public","enrichment_json":"{\n  \"version\": \"2.0\",\n  \"pipeline_stage\": \"enriched\",\n  \"title\": \"FRL Discrete Gestures: Hand Gesture Recognition from Surface Electromyography\",\n  \"description\": \"This dataset comprises wrist-based surface electromyography (sEMG) recordings from 100 participants performing nine discrete hand gestures (four thumb swipes, four finger-to-thumb pinches, and one thumb tap) for gesture-based human-computer interaction. Recorded at 2000 Hz using a 16-channel dry electrode wristband, the dataset enables development and evaluation of generic machine learning models for real-time gesture classification without user calibration. The data supports research in neuromotor interfaces for AR/VR applications, accessibility, and alternative input modalities.\",\n  \"methods_description\": \"Participants wore a single-wristband sEMG research device (16 channels, 2000 Hz sampling rate, 12-bit resolution, ±6.6 mV dynamic range, 20-850 Hz bandwidth) on the dominant wrist. Gestures were prompted via visual cues in randomized order with variable inter-gesture intervals. Each session included multiple repetitions of the nine gesture types across 16 recording stages, yielding approximately 1900 total prompted gestures. Signal preprocessing included 40 Hz high-pass filtering and clock drift correction.\",\n  \"license\": \"CC-BY-NC 4.0\",\n  \"authors\": {\n    \"Patrick Kaifosh\": {},\n    \"Thomas R. Reardon\": {},\n    \"CTRL-labs at Reality Labs\": {}\n  },\n  \"keywords\": [\n    {\n      \"term\": \"Electromyography\",\n      \"subject_scheme\": \"MeSH\",\n      \"value_uri\": \"http://id.nlm.nih.gov/mesh/D004576\"\n    },\n    {\n      \"term\": \"gesture recognition\"\n    },\n    {\n      \"term\": \"human-computer interaction\"\n    },\n    {\n      \"term\": \"machine learning\"\n    },\n    {\n      \"term\": \"motor control\"\n    },\n    {\n      \"term\": \"wearable sensors\"\n    },\n    {\n      \"term\": \"real-time classification\"\n    }\n  ],\n  \"related_identifiers\": [\n    {\n      \"identifier\": \"10.1038/s41586-025-09255-w\",\n      \"identifier_type\": \"DOI\",\n      \"relation_type\": \"References\"\n    },\n    {\n      \"identifier\": \"https://github.com/nemarDatasets/nm000105\",\n      \"identifier_type\": \"URL\",\n      \"relation_type\": \"IsDescribedBy\"\n    },\n    {\n      \"identifier\": \"https://nemar.org/dataexplorer/detail?dataset_id=nm000105\",\n      \"identifier_type\": \"URL\",\n      \"relation_type\": \"IsDescribedBy\"\n    },\n    {\n      \"identifier\": \"10.5281/zenodo.17613958\",\n      \"identifier_type\": \"DOI\",\n      \"relation_type\": \"IsIdenticalTo\"\n    },\n    {\n      \"identifier\": \"10.82901/nemar.nm000105\",\n      \"identifier_type\": \"DOI\",\n      \"relation_type\": \"IsVersionOf\"\n    }\n  ],\n  \"funding_references\": [\n    {\n      \"funder_name\": \"Meta Reality Labs\"\n    }\n  ],\n  \"resource_type_specific\": \"EMG Dataset\",\n  \"modalities\": [\n    \"emg\"\n  ],\n  \"source_hash\": \"765b53c836ebd1dc397c3bc8a947a223469870b3ce5ec6ae1ec0824fe049fead\"\n}","last_activity_at":"2026-02-12 09:49:28","source":null,"source_id":null,"subject_count":100,"modalities":"emg","age_min":null,"age_max":null,"file_size":22107418852,"total_files":512,"tasks":"discretegestures","metadata_columns_error":null,"staleness_warn_stage":null,"staleness_admin_notified_at":null,"authors":"Patrick Kaifosh, Thomas R. Reardon, CTRL-labs at Reality Labs","license":"CC-BY-NC 4.0","readme":"# discrete_gestures: Discrete Hand Gesture Detection from EMG\n\n## Overview\n\n**Dataset**: discrete_gestures - Discrete hand gestures from wrist-based surface electromyography\n**Task**: Nine discrete hand gestures (pinches and swipes)\n**Participants**: 100 subjects\n**Sessions**: 100 total (1 per subject)\n**Publication**: Kaifosh et al., 2025 - \"A generic non-invasive neuromotor interface for human-computer interaction\" (Nature)\n\n### Purpose\n\nThis dataset captures wrist-based sEMG signals during prompted discrete hand gestures for navigation and activation tasks. The goal is to enable gesture-based computer control without cameras or visible hand movements, with applications in AR/VR, mobile interfaces, and accessibility.\n\nKey research objectives:\n- Generic models that work across users without calibration\n- Discrete gesture classification with high accuracy\n- Real-time gesture detection for interactive systems\n- Robustness to electrode placement variability\n\n## Dataset Details\n\n### Participants\n\n**Sample size**: 100 participants\n**Demographics**: Not available (age, sex, handedness marked as n/a)\n**Recording side**: Dominant wrist (assumed right-handed, varies by participant)\n**Sessions**: 1 session per participant\n\n### Hardware\n\n**Device**: sEMG Research Device (sEMG-RD)\n**Configuration**: Single wristband (dominant wrist)\n**Channels**: 16\n**Sampling rate**: 2000 Hz\n**Bit depth**: 12 bits\n**Dynamic range**: ±6.6 mV\n**Bandwidth**: 20-850 Hz\n**Connectivity**: Bluetooth\n**Electrode type**: Dry gold-plated differential pairs\n\n### Gestures\n\n**Nine discrete gestures**:\n\n**Thumb swipes** (4):\n- Left swipe\n- Right swipe\n- Up swipe\n- Down swipe\n\n**Pinches** (4):\n- Index-to-thumb pinch\n- Middle-to-thumb pinch\n- Ring-to-thumb pinch\n- Pinky-to-thumb pinch\n\n**Activation** (1):\n- Thumb tap\n\n### Recording Protocol\n\n1. Participant dons sEMG-RD on dominant wrist\n2. Gesture prompter displays gesture cue (scrolling left-to-right)\n3. Participant performs prompted gesture\n4. Randomized order with randomized inter-gesture intervals\n5. Multiple repetitions of each gesture type\n\n**Session duration**: Varies by participant\n**Total gestures**: 1900 prompted gestures across all participants\n**Stage boundaries**: 16 recording stages per session\n\n## Data Contents\n\n### Files per Session\n\n```\nsub-XXX/ses-XXX/emg/\n├── sub-XXX_ses-XXX_task-discretegestures_emg.edf\n├── sub-XXX_ses-XXX_task-discretegestures_emg.json\n├── sub-XXX_ses-XXX_task-discretegestures_channels.tsv\n├── sub-XXX_ses-XXX_task-discretegestures_events.tsv\n└── sub-XXX_ses-XXX_electrodes.tsv\n```\n\n### Channel Configuration\n\n**Total channels**: 16 (EMG0-EMG15)\n**Channel naming**: Unique identifiers (EMG0-EMG15)\n**Electrode naming**: E0-E15 (physical positions)\n**Reference**: Bipolar (differential sensing)\n\n**channels.tsv columns**:\n- `name`: Channel identifier (EMG0-EMG15)\n- `type`: EMG\n- `units`: V\n- `signal_electrode`: Physical electrode name (E0-E15)\n- `reference`: bipolar\n\n**electrodes.tsv columns**:\n- `name`: Electrode identifier (E0-E15)\n- `x`, `y`, `z`: 3D coordinates (percent units, no decimals)\n\n### Events\n\n**events.tsv contains**:\n- **Gesture prompts**: Timestamped prompts for each gesture\n  - `type`: gesture_X (where X is the gesture name)\n  - `latency`: Sample index when gesture was prompted\n  - `gesture_type`: Specific gesture (e.g., \"index_pinch\", \"thumb_swipe_left\")\n- **Stage boundaries**: Recording session phases\n  - `type`: stage_boundary\n  - `stage_name`: Stage identifier\n\n**Total events**: 1916 (1900 gesture prompts + 16 stage boundaries)\n\n### Coordinate System\n\n**Single coordinate system** (no space entity):\n\n```\nEMGCoordinateSystem: Other\nEMGCoordinateUnits: percent\nX: USP → RSP (0-100%)\nY: Right-hand rule perpendicular (0-100%)\nZ: Radial offset (constant 10%)\n```\n\n**Anatomical landmarks**:\n- RSP: Radial Styloid Process\n- USP: Ulnar Styloid Process\n\n**Note**: Right-handed coordinate system for dominant wrist\n\n## Signal Processing\n\n### Preprocessing Applied\n\n1. **High-pass filtering**: 40 Hz cutoff\n2. **Clock drift correction**: Time synchronization\n3. **Irregular sampling handling**: Resampling when deviation >1% (up to 9290% deviation detected)\n\n### Signal Characteristics\n\n**Gesture patterns**:\n- Patterned activity across channels corresponding to flexor/extensor muscles\n- Fine differences across gesture instances\n- Channel activity correlates with muscle positions (Fig. 1 in paper)\n\n## Baseline Performance\n\n### Published Results (Kaifosh et al., 2025)\n\n**Offline Classification** (held-out participants):\n- Accuracy: >90% for gesture classification\n- False-negative rate improves with more training data\n- Generic models trained on hundreds of participants\n\n**Closed-loop Performance** (n=24 naive test users):\n- **First-hit probability**: Median improvement from 0.74 (practice) to 0.82 (evaluation block 2)\n- **Gesture completion rate**: Median 0.88 gestures/second (evaluation block 2)\n- **Baseline comparison**: Gaming controller achieves 1.45 completions/second\n\n**Model architecture**: 1D convolution → LSTM layers\n**Learning effects**: Participants improve from practice to evaluation blocks\n\n### Representation Analysis\n\n**Network learns**:\n- First layer filters resemble motor unit action potentials (MUAPs)\n- Deeper layers progressively separate gesture categories\n- Invariance to nuisance variables (participant ID, electrode placement, signal power)\n\n## Confusion Matrix\n\n**Common confusions** (from paper):\n- Index and middle holds sometimes released too early\n- Similar gestures (e.g., adjacent finger pinches) occasionally confused\n- Swipe directions generally well-separated\n\n**Note**: Some errors are behavioral (wrong gesture performed) not just decoding errors\n\n## Use Cases\n\n### Machine Learning\n\n- **Time series classification**: Discrete event detection\n- **Generic modeling**: Out-of-the-box cross-user generalization\n- **Representation learning**: Physiologically-grounded features\n- **Real-time prediction**: Low-latency gesture detection\n\n### Applications\n\n- **Grid navigation**: Discrete movement in 2D space\n- **Menu selection**: Activation gestures for UI elements\n- **Game control**: Gesture-based game inputs\n- **AR/VR interfaces**: Hands-free navigation\n- **Accessibility**: Alternative input modality\n\n## Known Issues and Limitations\n\n### By Design\n\n- **Single wrist**: Dominant hand only (not bilateral)\n- **Handedness unknown**: Assumed right-handed, varies by participant\n- **Gesture novelty**: Users needed coaching to learn effective gestures\n- **No demographic data**: Age, sex, handedness not collected\n\n### Technical\n\n- **Electrode placement**: Single session per user (less cross-session data than emg2qwerty)\n- **Signal amplitude**: Varies with gesture force\n- **Hardware unavailable**: sEMG-RD not commercially available\n\n### Data Quality\n\n- **Irregular sampling**: High deviation detected (up to 9290%), resampling applied\n- **Behavioral errors**: Not all errors are decoder errors (some user mistakes)\n\n## Comparison to Baselines\n\n**Nintendo Joy-Con controller**:\n- Median: 1.45 completions/second\n- sEMG decoder: 0.88 completions/second (66% slower)\n\n**However**: sEMG doesn't require hand-encumbering device\n\n### BIDS Format\n\n```\nPernet, C.R., et al. (2019). EEG-BIDS, an extension to the brain\nimaging data structure for electroencephalography.\nScientific Data, 6(1), 103.\n```\n\n## Access and Contact\n\n**Original data**: Part of Meta Reality Labs neuromotor interface research\n**BIDS conversion**: Custom MATLAB tools using EEGLAB BIDS plugin\n**Data curator**: Yahya Shirazi, SCCN (Swartz Center for Computational Neuroscience), INC (Institute for Neural Computation), UCSD\n**Contact**: See Nature paper for corresponding authors\n\n## License\n\nResearch and educational use. See original publication.\n\n\n## Citation\n```\nKaifosh, P., Reardon, T.R., & CTRL-labs at Reality Labs. (2025).\nA generic non-invasive neuromotor interface for human-computer interaction.\nNature, 645(8081), 702-711. https://doi.org/10.1038/s41586-025-09255-w\n```\n\n## Data Curator\n**Yahya Shirazi**\nSCCN (Swartz Center for Computational Neuroscience)\nINC (Institute for Neural Computation)\nUniversity of California San Diego\n\n## Version Histo","bids_version":"1.11.0","sessions_count":1,"publish_date":"2026-01-19 03:25:23","embedding_dirty":0,"license_tier":"noncommercial","zarr_status":"ready","zarr_converted_at":"2026-09-05 03:19:24","zarr_store_count":100,"zarr_index_etag":"a62ba986cdc5c8e173efa71bdca27623","zarr_source_commit":"bf4c56fc66c30f1c4d6a6c476ec89dcb3a0b11c9","archive_status":"ready","archive_size":10606554300,"archive_retry_count":0,"records_status":null,"archive_skip_reason":null,"zarr_errors":0,"zarr_failure_count":0,"zarr_deterministic":0,"zarr_failed_at":null,"num_dataset_citations":0,"num_datapaper_citations":0,"n_channels":null,"electrode_system":null,"has_hed":0,"hed_version":null,"is_exemplar":0,"bytes_present":36836870436,"data_complete":1,"withdrawn_at":null,"withdrawn_reason":null,"archive_complete":null,"archive_absent_files":null,"archive_declared_files":null,"zarr_pool_breaks":0,"total_recording_duration":230225,"recording_duration_min":1532,"recording_duration_max":2966,"recording_count":100,"recordings_unavailable":0,"recordings_measured":100,"channel_count_min":16,"channel_count_max":16,"sampling_frequency":null,"power_line_frequency":null,"eeg_reference":null,"placement_scheme":null,"sweep_stamps":"{\"enrichment_updated_at\":\"2026-02-23 07:49:12\",\"metadata_updated_at\":\"2026-05-16 07:29:39\",\"archive_checked_at\":\"2026-06-05 01:32:45\",\"zarr_checked_at\":\"2026-06-07 17:57:59\",\"records_checked_at\":null,\"citations_updated_at\":null,\"channel_montage_checked_at\":null,\"hed_checked_at\":\"2026-06-30 03:44:49\",\"data_checked_at\":\"2026-07-22 04:44:25\",\"availability_report_at\":\"2026-07-23 01:04:12\",\"signal_defaults_at\":\"2026-09-02 11:34:42\",\"zarr_verify_attempted_at\":\"2026-09-04 03:01:10\",\"zarr_verified_at\":\"2026-09-04 03:01:10\",\"zarr_verified_commit\":\"bf4c56fc66c30f1c4d6a6c476ec89dcb3a0b11c9\",\"zarr_verify_status\":\"verified\",\"zarr_verify_examples\":[],\"zarr_verify_sampled\":40.0,\"zarr_verify_checked\":40.0,\"zarr_verify_checked_channels\":40.0,\"zarr_verify_checked_duration\":40.0,\"zarr_verify_checked_rate\":40.0,\"zarr_verify_unchecked\":0.0,\"zarr_verify_mismatch_count\":0.0,\"zarr_verify_examples_truncated\":0.0,\"recording_stats_at\":\"2026-09-06 03:00:29\"}","participants":100,"num_citations":0,"latest_version":"v2.0.0","zarr_verify_status":"verified","zarr_verified_at":"2026-09-04 03:01:10","owner_username":"yahya","owner_github":"neuromechanist","file_size_formatted":"20.59 GB","zarr_data_failures":null,"zarr_index_url":"https://zarr.nemar.org/nm000105/zarr/index.json","attestation_deposit_type":null,"attestation_key_status":null,"attestation_deidentified":null,"attestation_no_duplicate":null,"attestation_upstream_source":null,"attestation_accepted_at":null}}