# MI-005: Noninvasive electroencephalogram based control of a robotic arm for reach and grasp tasks > Generated from a local paper card. Do not edit this page directly; edit the source card and rerun `vp run docs:generate`. > Internal wiki boundary: paper-card evidence and user-provided PDF downloads are available for private reading. Extracted full-paper text is not published. > Source: `library/paper_cards/MI-005.md` ## Paper Access * Internal PDF: Download Paper * DOI / official page: [10.1038/srep38565](https://doi.org/10.1038/srep38565) * Deployment boundary: these PDF links are intended only for a private/protected internal wiki. ## MI-005: Noninvasive electroencephalogram based control of a robotic arm for reach and grasp tasks ## Metadata * ID: MI-005 * Title: Noninvasive electroencephalogram based control of a robotic arm for reach and grasp tasks * Year: 2016 * DOI / URL: 10.1038/srep38565 * Local PDF: see Paper Access section above * Text artifact: local-only path withheld from docs site * Review status: `extracted` ## Study Type * Track: Motor Imagery * Task: noninvasive EEG motor-imagery control of robotic-arm reach and grasp tasks * Participants or dataset: 18 healthy participants recruited; 5 exited early; 13 analyzed; 7 female; mean age 27.3; 8-15 sessions over about 81 +/- 34 days * Hardware: 64-channel Neuroscan cap with SynAmps RT; JACO 7-DOF three-finger robotic arm; Microsoft Kinect for target localization * Channels or sensors: 62 EEG channels acquired; online control used C3/C4 and neighboring motor-cortex channels ## Methods * Paradigm: left-hand, right-hand, both-hands motor imagery and relaxation mapped to left/right/up/down control; reach and grasp decomposed into 2D hover selection plus 1D downward grasp * Signal processing or model: upper mu band 10-14 Hz; small Laplacian spatial filter; 16th-order autoregressive model; 400 ms window estimating mu amplitude; linear mapping from mu power to cursor or arm velocity * Training/calibration: multi-session progression from virtual cursor to four-target, five-target, random-target, and shelf-target grasp tasks * Online/offline: online BCI robotic-arm control with post-hoc ERD/ERS analysis ## Results * Metrics: 1D left-right PVC improved from 78.4 +/- 7.0% to 90.2 +/- 3.1% by the second session and later exceeded 95%; 2D cursor performance exceeded 85% after seven sessions; four-target grasp PVC 80.3 +/- 17.0%; five-target grasp PVC 77.9 +/- 14.7% * Main findings: healthy participants could use noninvasive EEG motor imagery to perform staged robotic reach-and-grasp tasks, but control was decomposed into low-dimensional sequential commands * Reported limitations: not fluid 3D continuous control; high-accuracy 3D control from at least three independent noninvasive EEG signals was not shown; hover confirmation reduces false selection but adds time; performance varies with user state and electrode placement ## Relevance To This Project * Supports: noninvasive EEG can drive robotic reach/grasp at a high-level task interface when the control problem is decomposed * Conflicts with: the paper does not support direct high-bandwidth 6DOF continuous arm control from noninvasive EEG * Design implication: use MI for mode/intervention/confirmation-style control and rely on scene perception/shared autonomy for low-level grasp execution ## Extracted Evidence | Claim | Status | Evidence Note | Page/Section | | --- | --- | --- | --- | | Noninvasive EEG motor imagery can support robotic-arm reach and grasp when the task is decomposed. | verified | The study reports multi-stage online control of a JACO arm using EEG motor imagery and hover-based target confirmation. | Abstract; Results; Methods | | Low-dimensional sequential MI control is not equivalent to fluid 3D continuous arm control. | verified | The discussion notes that high-accuracy 3D continuous control with three independent noninvasive EEG signals had not been demonstrated. | Discussion | | Hover or dwell confirmation can reduce unintended target selection at the cost of time. | verified | The protocol used a 2 s hover confirmation before target selection in grasp tasks. | Methods / task protocol | | SAH-BRI-Grasp should use MI as an intervention or mode-control signal rather than the primary low-level arm controller. | inferred | This follows from the paper's successful but low-dimensional sequential control design and its stated continuous-control limitations. | Results; Discussion | ## Open Questions * How much MI training is acceptable for the intended system? * Should MI be used for cancel/stop, mode switch, or execute confirmation in the first prototype? * Can MI be combined with SSVEP without increasing fatigue or false activations?