The trip succeeds if episodes land on the SSD and ingest runs on the laptop. Hotel Wi-Fi is optional.
- Install the capture app:
cd apps/mobile && npx expo start(scan with Expo Go) or a TestFlight/dev build. - Install ingest on the laptop:
python3 -m venv .venv
source .venv/bin/activate
pip install -r apps/ingest/requirements.txt- Pack the kit in hardware.md. Chest harness first. Handheld last.
- Get permission. No secret factory filming. No children in frame.
- Open the app. Accept consent (EN/ZH). You must be 18+.
- Pick a Shenzhen campaign chip. Confirm the why sentence.
- Chest-mount the phone. Both hands free. Start record.
- Stop. Keep. Repeat.
- Library → Export to Files / USB onto the SSD. Do this every few episodes.
Each episode folder must contain:
episode.json
intent.json
device.json
imu.jsonl
video.mp4
source .venv/bin/activate
pip install -r apps/ingest/requirements.txt
export PYTHONPATH=apps/ingest
python -m dvidia_ingest ingest /Volumes/SSD/ep_XXXX --out ./data/processed --synthetic
# real 21-point hands (downloads the MediaPipe task file):
# pip install -r apps/ingest/requirements-mediapipe.txt
# python -m dvidia_ingest ingest /Volumes/SSD/ep_XXXX --out ./data/processedAccepted packages land at data/processed/<skill-slug>/ as LeRobot v3.
Copy into the store:
python -m dvidia_ingest copy-into-web ./data/processed/electronics.cable-dress
pnpm --filter @dvidia/web devA lab installs with:
curl -X POST http://localhost:3000/api/v1/skills/electronics.cable-dress/install \
-H "Authorization: Bearer sk_demo_dvidia_local"- Duration ≥ 4s
- Hands visible on ≥ 60% of frames
- Not a pocket / dark video
- Intent is a real why, not a blank
- Consent version
2026-08-30.v1
Handheld episodes are kept but marked lower-tier.