# BRI-005: Reach and grasp by people with tetraplegia using a neurally controlled robotic arm > 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/BRI-005.md` ## Paper Access * Internal PDF: Download Paper * DOI / official page: [10.1038/nature11076](https://doi.org/10.1038/nature11076) * Deployment boundary: these PDF links are intended only for a private/protected internal wiki. ## BRI-005: Reach and grasp by people with tetraplegia using a neurally controlled robotic arm ## Metadata * ID: BRI-005 * Title: Reach and grasp by people with tetraplegia using a neurally controlled robotic arm * Year: 2012 * DOI / URL: 10.1038/nature11076 * Local PDF: see Paper Access section above * Text artifact: local-only path withheld from docs site * Review status: `extracted` ## Study Type * Track: Intracortical brain-robot interface / robotic reach and grasp * Task: BrainGate neural interface control of human-scale robotic arms for 3D reach-and-grasp targets and one bottle-drinking assistive task * Participants or dataset: 2 BrainGate2 participants with long-standing tetraplegia and anarthria after brainstem stroke; S3 was a 58-year-old woman and T2 was a 65-year-old man * Hardware: BrainGate Neural Interface System; DLR Light-Weight Robot III with DLR Five-Finger Hand; DEKA Generation 2 prosthetic arm; automated servo-based 3D target platform * Channels or sensors: 4 x 4 mm, 96-channel intracortical microelectrode array implanted in dominant motor-cortex arm/hand area; extracellular unit threshold crossings from neural signals sampled at 30 kHz ## Methods * Paradigm: continuous user-driven 3D endpoint velocity control plus a simultaneously decoded grasp state; foam-ball targets were presented one at a time in 3D space, and S3 also performed a 2D tabletop bottle-drinking task with sequential hand actions * Signal processing or model: threshold crossing rates were used as features; intended endpoint velocity was decoded with a Kalman filter; intended hand state was decoded with linear discriminant analysis; T2 also used common-average referencing and velocity-bias correction * Training/calibration: each session began with one open-loop calibration block followed by several closed-loop calibration blocks with decreasing error attenuation; average decoder calibration was about 31 minutes, excluding time between blocks * Online/offline: online intracortical BRI control in a pilot clinical-trial setting ## Results * Metrics: target touch rate, target grasp rate, time to touch, time to grasp, and bottle-drinking attempts * Main findings: S3 touched targets in 48.8% of DLR trials and 69.2% of DEKA trials, and grasped targets in 21.3% and 46.2% of all trials respectively; T2 touched targets in 95.6% of DEKA trials and grasped targets in 62.2% of all trials; all sessions were above chance; S3 completed the bottle-drinking task in 4 of 6 attempts after brief familiarization * Reported limitations: control was invasive and based on intracortical spiking rather than EEG; only two participants were studied; hand orientation was not under user control in the 3D task; target scoring relied on visual inspection; the bottle task used 2D velocity control plus automated sequential hand actions; the paper states that future work is needed for more signals, better decoders, explicit training, and more natural control ## Relevance To This Project * Supports: a strong invasive comparator showing that motor-cortex neural activity can drive robotic reach and grasp for people with tetraplegia * Conflicts with: SAH-BRI-Grasp is noninvasive and scene-aware, whereas this paper uses intracortical arrays, no SSVEP, no EEG-based MI, no YOLO perception, and no visual dynamic command space * Design implication: use BRI-005 as an upper-bound/direct-control comparator and avoid implying that noninvasive EEG should reproduce intracortical continuous endpoint-velocity control ## Extracted Evidence | Claim | Status | Evidence Note | Page/Section | | --- | --- | --- | --- | | Two people with long-standing tetraplegia used intracortical motor-cortex signals to control robotic reach and grasp. | verified | The participants S3 and T2 had tetraplegia and anarthria after brainstem stroke, and neural signals came from implanted 96-channel arrays in motor cortex. | Abstract; Main text; Methods | | The online control architecture decoded continuous 3D endpoint velocity and a separate hand action state. | verified | The paper describes Kalman-filter decoding for endpoint velocity and linear-discriminant decoding for hand state from threshold crossing rates. | Main text; Methods, Filter calibration; Methods, Hand velocity and grasp filters | | The 3D task required more than simple point-to-point reaching. | verified | The target setup required alignment with small foam balls, avoidance of support rods, curved paths, and corrective actions; success was judged from video inspection. | Main text; Methods Summary; Target presentation | | BRI-005 demonstrated above-chance target contact and grasp performance in all reported robot sessions. | verified | S3 touched 48.8% of DLR and 69.2% of DEKA targets, while T2 touched 95.6% of DEKA targets; all sessions were reported as above chance. | Results; Table 1 | | The bottle-drinking task shows limited ADL-oriented assistive use, but with robot-side sequencing. | verified | S3 completed 4 of 6 bottle-drinking attempts; the task restricted velocity to 2D and used decoded grasp state to trigger one of four phase-dependent hand or arm actions. | Results; Methods, Sequential activation of DLR robot hand actions | | BRI-005 is best treated as an invasive direct-control comparator for SAH-BRI-Grasp. | inferred | The paper supports robotic reach-and-grasp from neural intent, but its intracortical continuous-control design differs from the repository's noninvasive scene-aware SSVEP-MI shared-autonomy direction. | Full paper; README-001 | | Transfer from BRI-005 performance to low-channel noninvasive EEG is unresolved. | needs confirmation | The local text does not test EEG, SSVEP, MI EEG decoding, YOLO-generated targets, or low-channel wearable configurations. | Full paper | ## Open Questions * Which comparator metrics should SAH-BRI-Grasp mirror: touch rate, grasp rate, time to touch, time to grasp, or ADL-style task completion? * How should the manuscript distinguish intracortical continuous-control results from noninvasive SSVEP-MI high-level intent control? * What baseline is fair for comparing scene-aware shared autonomy against direct endpoint-velocity control when sensing, invasiveness, and autonomy differ? * The local text does not provide evidence for dynamic visual target generation, workload, or noninvasive decoding transfer.