# 实验路线图

> 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: `research/04-experiment-roadmap.md`

## Experiment Roadmap

Status: `pre-data roadmap`

The experiments are designed to turn literature-supported design rationale into project-specific evidence.

## Experiment Chain

```text
Exp1: Can scene objects become SSVEP-selectable targets?
  -> Exp2: Can MI control mode/intervention states reliably?
  -> Exp3: Can the complete shared-autonomy loop perform grasping?
  -> Exp4: Which EEG channels are necessary for low-channel translation?
```

## Experiment-To-Claim Map

| Experiment                       | Primary Claim                                                             | Current Status            |
| -------------------------------- | ------------------------------------------------------------------------- | ------------------------- |
| Exp1 SSVEP-YOLO Target Selection | Visual scene candidates can become BCI-selectable semantic targets.       | `pre-data protocol draft` |
| Exp2 MI Mode Control             | MI can support intervention, stop, and mode-control states.               | `pre-data protocol draft` |
| Exp3 Closed-Loop Grasping        | High-level EEG intent plus shared autonomy can complete robotic grasping. | `pre-data protocol draft` |
| Exp4 Channel Ablation            | Low-channel EEG configurations can preserve useful task performance.      | `pre-data protocol draft` |

## Required Records

Each evidence-producing session must bind:

* `session_id`
* `hardware_inventory_id`
* `threshold_freeze_record_id`
* `calibration_record_id` when robot/camera calibration is involved
* runtime JSONL path
* participant/session inclusion status
* trial-level failure modes

## Result Claim Gate

Results, workload, grasp success, task completion time, false activation, channel ablation, and product-readiness claims remain blocked until data, exclusions, confidence intervals, and failure-mode logs exist.