# One-Pager: FIELD Lane — AR Glasses for Cherry Pick Readiness

**For:** James Grunsky + parallel AI workstreams  
**Lane owner (this doc):** FIELD — ripeness cues, <1 s HUD, phone → wearable  
**Out of lane:** SO-ARM101 robot arm (handoff only later)  
**Full brief:** `/workspace/ar-cherry-glasses/FIELD_BRIEF.md`  
**Date:** 2026-09-04

---

## What we’re building

Hands-free (eventually) guidance so pickers in **Lodi / Stockton sweet cherry** orchards get a fast **READY / WAIT / MAYBE** cue on fruit color maturity — improving packout consistency under labor pressure. **Not** a sugar meter; **not** the robot yet.

## How CA cherries are actually judged

- **Primary field signal:** skin color (light red → mahogany / dark mahogany for dark sweets; separate rules for **Rainier** blush).
- **Legal/quality companion:** ~**14–16% Brix** minimum by variety (UC Davis / CA practice); premium often wants fuller color + higher SSC.
- **Also matters, camera can’t measure:** firmness, taste (SSC/TA), internal quality.
- **Local varieties to plan for:** Bing, Coral, Brooks, Rainier, Lapin, Tulare (confirm Handel Road mix).
- Ethylene does **not** ripen cherries after pick — get color right in the orchard.

## What the glasses must show (<~1 s)

| Badge | Meaning |
| --- | --- |
| READY | High-confidence pick for this variety/block color standard |
| WAIT | Too light / immature — leave |
| MAYBE | Borderline or occluded / bad light — human judgment |

Icons + bilingual EN/ES; peripheral monocular HUD; prefer **high precision on READY** (avoid picking too early). Fail soft if slow — no stale labels.

## Build path

| Phase | Deliverable | Success (rough) | Effort (OOM) |
| --- | --- | --- | --- |
| **A Phone** | Orchard app, offline model, label mode | READY precision ~≥90% vs QC; <1 s; minimal slowdown | ~4–10 person-weeks + few half-days in orchard |
| **B Glasses** | RealWear-class assisted reality **or** bright display glasses | Shift-usable; metrics hold; crew OK | ~8–20 person-weeks post-A; ~$0.8–3k/unit |
| **C Robot handoff** | Shared labeled ontology + detections for later vision | Reusable dataset/checkpoint | Package after A data exists |

**Hardware honesty:** Phone first. **RealWear Navigator ~$2.6k / IP66** fits dusty heat; **Meta Ray-Ban / Display (~$300–800)** better for capture/light HUD than all-day dust; XREAL-class weak for rugged outdoor shifts.

## Adjacent tech (don’t reinvent blind)

- Cherry maturity CV in orchards (e.g. YOLOX-based sweet cherry work): doi.org/10.3390/agronomy12102482  
- Vine **AR pruning** on Vuzix: 3d2cut.com/ai-ar-pruning-glasses/  
- **Agerpoint + Meta** wearable AI pilots for CA permanent crops (2026 grant news)

## Data without slowing harvest

Scout/QC photo passes; morning **20 fruit/variety photo+Brix**; packer color-sorted samples as labels; expand egocentric images across sun/shade. Start hundreds of good labels/variety, grow toward thousands.

## Ask James (top 3)

1. Year-one cultivars: Bing only or Bing+Coral+Rainier?  
2. Packer color card + target: legal minimum vs premium mahogany/Brix? Confirm **Morada** current row cards (provisional READY size ~10-row / 26.6 mm — see `/docs/morada-sizing`).  
3. Cameras/glasses OK with crew + Spanish-first silent HUD?

**Pack house locked:** Handel → **Morada Produce (Linden)** / **UNITEC Cherry Vision** (not TOMRA). Color primary; size secondary.

## Other AIs: please don’t collide

- **FIELD (this lane):** decision model, UX timing, phone MVP, orchard data protocol.  
- **Hardware / robot lanes:** assume Phase A labels + READY ontology as the shared interface; don’t block on glasses SKU until phone metrics exist.

**Next action:** Phase A phone prototype at first color-break — one variety, one trellis row, timed trial vs veteran picker.