# Decision Log > Generated from repository source files. Do not edit this page directly; edit the source file listed below and rerun `vp run docs:generate`. > Source: `research/07-decision-log.md` ## Decision Log Status: `wiki design rationale` This page records major design decisions behind `SAH-BRI-Grasp`. It is not an experiment result log. ## Decision 1: Frame EEG As High-Level Intent Decision: non-invasive EEG should not be used as dense low-level robot-arm control in this project. Rationale: BCI foundation and MI evidence support strict evaluation boundaries, calibration limits, online validation needs, and low-bandwidth control constraints. The project therefore treats EEG as a high-level intention and intervention channel. Status: `inferred` Implication: the paper should avoid claiming direct continuous 6-DoF EEG arm control. ## Decision 2: Assign SSVEP To Target Selection Decision: SSVEP is the main discrete visual target-selection channel. Rationale: SSVEP literature supports frequency-tagged selection, high-speed fixed-target decoders, benchmark datasets, and recent short-window/deep-learning trends. The unresolved transfer is from controlled fixed targets to scene-bound object boxes. Status: `inferred` Implication: Exp1 must test fixed, frozen-scene, and live-scene candidate conditions. ## Decision 3: Assign MI To Mode Control And Intervention Decision: MI is scoped to active mode control, confirmation, cancel, pause, stop, or intervention. Rationale: MI literature supports motor-imagery decoding and some robot-control comparators, but false activation, no-control rejection, latency, and training burden remain major constraints. Status: `inferred` Implication: Exp2 must include rest/no-control evaluation before MI can be used as a safety-relevant online command. ## Decision 4: Treat YOLO As Command-Space Generator Decision: YOLO-style detection is used to generate selectable object candidates, not to prove grasp execution. Rationale: detection papers support real-time object candidate generation, while grasping literature separates detection from grasp-pose estimation, calibration, reachability, and execution. Status: `inferred` for command-space use; `verified` for the detection-vs-grasp boundary. Implication: Exp1 should log candidate jitter, target loss, and class confusion; Exp3 should log grasp-pose, calibration, planning, and execution failures separately. ## Decision 5: Use Shared Autonomy As The Execution Layer Decision: shared autonomy arbitrates between uncertain user intent and robot-side feasibility. Rationale: shared-control literature supports confidence-aware assistance and policy blending, while BRI comparators show that robot autonomy can reduce the burden of low-dimensional human input. Status: `inferred` Implication: Exp3 should compare scene-aware shared autonomy against fixed command-space and single-modality baselines. ## Decision 6: Keep Results Blocked Until Experiments Exist Decision: Results, result-dependent Discussion, and Conclusion remain placeholders. Rationale: no local Exp1-Exp4 data exist yet, and the evidence policy forbids unsupported system-performance claims. Status: `verified` as current project state. Implication: the docs site can guide experiments and writing, but it must not present final grasp success, workload, or low-channel performance claims. ## Decision 7: Keep Product Directions Separate From Scientific Claims Decision: product directions are strategic narratives, not verified outcomes. Rationale: the research paper output suite, embodied AI + BCI concept, and rehabilitation direction have different evidence requirements. Status: `inferred` for product positioning; `needs confirmation` for product-market or clinical benefit. Implication: product pages should stay useful for ideation without weakening the manuscript evidence boundary. ## Decision 8: Stage Dexterous Arm-Hand Work After The Current Baseline Decision: treat a self-designed arm plus open-source dexterous hand as a staged extension, not as a dependency for Exp1-Exp4 or the first SAH-BRI-Grasp claim set. Rationale: the research value comes from hierarchical object-task-skill selection, skill-level shared autonomy, and feedback-aware arm-hand coordination. Mounting a multi-finger hand alone does not verify these contributions and adds mechanical, electrical, calibration, control, and safety dependencies before the current closed-loop baseline exists. Status: `inferred` Implication: first keep the end effector interchangeable, validate a small primitive library through replay or scripted trials, and create a dedicated Exp5 only after repeatable local evidence exists. AmazingHand is one candidate engineering platform, not a pre-validated system choice. ## Decision 9: Keep SAH-BRI-Manip Provisional Until The Research Question Changes Decision: reserve `SAH-BRI-Manip` as a provisional working name for a future dexterous-manipulation system; do not rename the current `SAH-BRI-Grasp` system merely because it uses a dexterous hand. Rationale: an end-effector change alters the apparatus, while a theme change requires task intention, skill selection, contact adaptation, or dexterous manipulation to become a primary experimental variable. Keeping these scopes separate protects the interpretability of the first system paper. Status: `verified` as a project naming and scope decision; future system feasibility remains `needs confirmation`. Implication: keep primitive-grasp pilots under `SAH-BRI-Grasp`. Promote `SAH-BRI-Manip` only after a repeatable arm-hand platform, dedicated evidence lane, frozen experiment protocol, baselines, metrics, and failure taxonomy exist.