Neural Noise

Large EEG database of freewill reaching and grasping tasks for brain-machine interfaces

2 min read

  • A large electroencephalogram database of freewill reaching and grasping tasks supports decoder development and benchmarking for brain-machine interfaces.1
  • The Nature dataset has high reuse value for neural signal processing and BCI methods and is tier-1 for near-term algorithm work.1 1

Weekly enrichment (2026-07-20)

  • The underlying source is a Scientific Data (2025) Data Descriptor by Thapa, Boggess, and Bae (University of Kentucky) presenting a large EEG dataset of freewill reaching and grasping tasks for brain-machine interfaces.2 3
  • The dataset contains 6,808 trials across 49 recording sessions from 23 healthy young adults (8 female, 15 male; ages 18–24), who freely chose the target object at their own pace rather than following experimenter-timed cues.2
  • Each recording synchronized 31 EEG channels plus 4 EOG, 3 accelerometer, and 1 audio channel on a single amplifier (BrainVision Recorder, Brain Products); 21 subjects were sampled at 250 Hz while 2 subjects (sub-13 and sub-15) were sampled at 1000 Hz and later downsampled to 250 Hz.2
  • Simultaneous EOG recording supports research on ocular-artifact removal — the authors note only nine publicly available EEG datasets include EOG.2
  • The prior sole freewill upper-limb EEG dataset included only 2 subjects performing a key-press task; this reaching-and-grasping collection is substantially larger and mirrors real-world daily movements.2
  • Benchmark validation used a linear support-vector-machine (LSVM) movement-intention classifier on 0.5-second windows: peak average accuracy was 90.6 ± 6.7% for the 0.5–1 s window, all window cases exceeded 75%, and the first 2 seconds stayed predominantly above 80%.2
  • The data are distributed as Freewill_EEG_Reaching_Grasping.zip in EEG-BIDS structure via Figshare (doi:10.6084/m9.figshare.28632599), bundling raw EEG files, an example experiment video, and code to reproduce the reported results.3
  • For context, the earlier WAY-EEG-GAL dataset provided 3,936 grasp-and-lift trials from 12 participants recorded with 32 EEG channels at 500 Hz plus 5 EMG channels.4
  • For context, a study decoding natural reach-and-grasp actions across 45 participants reported grand-average peak accuracies of 62.3% (water-based), 56.4% (dry), and 61.3% (gel) electrode systems versus roughly 45.8% chance, underscoring the difficulty of noninvasive movement decoding.5

Footnotes

  1. https://news.google.com/rss/articles/CBMiX0FVX3lxTE1ISXlNSDk1MThVY1Q1VnFLZ2czWkZfekg1T3NwbGRxYWpaZTVucXY1ZDBKVEhzemZZUG1EN0tGVV9YclpTUzROUm43RGVTZ01QT3pLN2IzQVdwV2JFOEZN?oc=5 ↩ ↩2 ↩3

  2. https://doi.org/10.1038/s41597-025-06039-9 ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  3. https://doi.org/10.6084/m9.figshare.28632599.v1 ↩ ↩2

  4. https://doi.org/10.1038/sdata.2014.47 ↩

  5. https://doi.org/10.3389/fnins.2020.00849 ↩

Graph View