# 论文草稿：引言

> 本页由仓库源文件生成。请不要直接编辑本页；修改下方 source 文件后运行 `vp run docs:generate`。

> 中文支持说明：本页是中文站点镜像，保留论文标题、证据状态和来源字段的原始表述，避免未经核实的翻译改变证据边界。

> 来源: `paper/sections/01-introduction.md`

## Introduction

Status: `scaffold ready`

## Draft Skeleton

`SAH-BRI` is the proposed framework name for a Scene-Aware Hybrid Brain-Robot Interface. The framework combines three elements: visual scene understanding, hybrid noninvasive EEG-based intent decoding, and robot shared autonomy. The current system instance, `SAH-BRI-Grasp`, focuses this framework on robotic grasping, where scene objects become selectable task candidates and robot autonomy handles low-level planning and execution.

The motivation is evidence-gated. `BCI-001` supports the general framing of BCI as a limited-capacity, adaptive communication and control channel. `MI-005` shows that noninvasive motor-imagery EEG can support robotic reach-and-grasp tasks when control is decomposed into low-dimensional sequential commands, but it does not establish fluid high-accuracy 3D continuous control from noninvasive EEG. `BRI-002` shows that vision and shared autonomy can improve BCI manipulation, but its BCI channel is intracortical rather than noninvasive. Therefore, the current paper should present high-level EEG intent plus robot autonomy as a design rationale, not yet as a completed performance claim.

`SAH-BRI-Grasp` addresses a specific design question: how can noninvasive SSVEP-MI intent be connected to physical object manipulation without forcing EEG to act as a dense low-level robot controller? The current answer is an inferred system architecture: YOLO-style perception generates scene candidates; SSVEP selects or confirms visual targets; MI handles mode-level intervention such as confirm, cancel, pause, or stop; and shared autonomy transforms selected objects into robot actions through planning, grasp synthesis, and safety checks.

## Claim-Gated Contributions

| Contribution Candidate | Status | Evidence Basis | Writing Rule |
| --- | --- | --- | --- |
| Define the `SAH-BRI` framework and `SAH-BRI-Grasp` system instance. | verified | `README-001` | Can be stated as project scope. |
| Motivate EEG as high-level intent rather than low-level continuous arm control. | inferred | `BCI-001`, `MI-005`, `BRI-002` | Present as design rationale. |
| Use SSVEP for semantic scene-target selection. | inferred | `SSVEP-001`, `SSVEP-003`, `SSVEP-004` | Present as a testable module hypothesis. |
| Use MI for intervention and mode control. | inferred | `MI-005`, `BCI-001` | Present as scoped design choice. |
| Treat YOLO outputs as a dynamic command space. | inferred | `YOLO-001`, `README-001` | Present as system contribution hypothesis. |
| Compare against shared-control baselines. | verified | `SA-001`, `SA-002` | Can cite as baseline formalism. |

## Required Caution

The introduction must not claim that `SAH-BRI-Grasp` already improves task success, grasp success, workload, user preference, or low-channel EEG performance. Those claims require Exp1-Exp4 results. The introduction should instead end by stating the planned evaluation path: dynamic SSVEP target selection, MI mode control, closed-loop grasping, and low-channel ablation.

## Paragraph Plan

1. Define `SAH-BRI` and the current `SAH-BRI-Grasp` instance.
2. Explain the mismatch between low-bandwidth EEG and dense robotic-arm control.
3. Introduce scene-aware command generation from visual perception.
4. Introduce hybrid SSVEP-MI as high-level intent and intervention input.
5. Introduce shared autonomy as the bridge from selected object to physical grasp.
6. List evidence-gated contributions and planned evaluation.