# Exp02 MI 模式控制

> 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: `experiments/exp02-mi-mode-control/protocol.md`

## Experiment 2: MI Mode Control

Status: `pre-data protocol draft`

## Goal

Test whether MI can act as active intervention and mode control for shared autonomy.

## Command Progression

1. Binary: `intent vs rest` or `confirm vs cancel`.
2. Ternary: `confirm / cancel / rest`.
3. Four-state: `enter-control / confirm / pause-stop / cancel`.

## Metrics

* Classification accuracy.
* Command latency.
* False activation rate.
* Mode-switch success rate.
* Training time.
* No-control state rejection performance.

## Analysis Plan

See `../analysis-plan.md`. Complete `../pre-data-freeze-checklist.md` and a session record before collecting evidence-producing trials.

Required pre-session records:

* `session_id`;
* `hardware_inventory_id`;
* `threshold_freeze_record_id`;
* protocol version and software commit;
* EEG montage, preprocessing, and MI decoder ID;
* command-set definition and no-control/rest labeling rule;
* training/pilot/test split;
* randomization seed/file and block order.

Trial inclusion rule:

* Include every attempted command and no-control/rest trial.
* Keep low-confidence and rejected decoder outputs in the runtime log.
* Count accepted commands during rest/no-control toward false activation unless a pre-frozen artifact exclusion applies.

Primary endpoints:

* false activation during rest/no-control;
* successful mode transitions for requested commands.

Baseline interpretation:

* binary commands establish separability;
* ternary commands add no-control rejection;
* four-state commands approximate full system mode control.

Progression rule:

* move from binary to ternary and from ternary to four-state only after pre-set accuracy and false-activation criteria are met. Numeric thresholds are `needs confirmation`.
* progression decisions must cite threshold IDs from `experiments/templates/threshold-freeze-record.md`.

## Evidence Output

Results update `library/EVIDENCE_MATRIX.md` rows for MI role and shared autonomy.

***

> 来源: `experiments/exp02-mi-mode-control/metrics.md`

## Metrics

| Metric | Definition | Status |
| --- | --- | --- |
| accuracy | correct MI command classifications / attempted command trials, with command-set denominator reported | definition draft frozen |
| command latency | EEG window end or cue onset to accepted command, with timing origin reported | definition draft frozen; timing origin must be frozen per protocol |
| false activation rate | accepted command during rest/no-control trials / rest/no-control trials | definition draft frozen |
| mode-switch success rate | correct state transitions / requested transitions | definition draft frozen |
| training time | calibration duration per participant/session | definition draft frozen |

***

> 来源: `experiments/exp02-mi-mode-control/variables.md`

## Variables

| Factor | Levels | Notes |
| --- | --- | --- |
| command set | binary / ternary / four-state | increase only after stable pilot |
| channel set | C3/Cz/C4 and expanded motor-area channels | align with README channel plan |
| feature/model | CSP/FBCSP/EEGNet or selected baseline | choose after literature review |
| feedback | offline / online feedback | pilot offline first |
| no-control state | included / excluded | required for asynchronous safety |

## Freeze Requirements

* Command labels and rest/no-control labeling must be fixed before the session.
* Decoder training, validation, and held-out test split must be recorded before scoring evidence-producing trials.
* Channel sets must map to exact electrode labels in `experiments/hardware-inventory.md`.
* Online feedback sessions must be analyzed separately from offline-only sessions.