# Arm v2 — Project Plan

**Status:** Scheduled, not started. Begin after PARA paper camera-ready.
**Trigger:** Blender procedural CAD is the bottleneck for iteration speed; v1 arm works but mechanical/calibration ceiling is reached.
**Goal:** A parametric-CAD'd, easy-to-iterate, mobile-base-ready arm v2 that becomes the substrate for the next 2 years of robotics work — campus deployment, eventual home-robot experiments, follow-on papers.

This is the *deliberate next project* after the PARA paper ships. It is **not** parallel to TRI or paper closeout. Putting it on the calendar with a real start date so it stops being a vague "someday."

---

## Why this is the right next project

- The PARA paper proved the *method* with cobbled-together hardware. The next paper(s) need a *dependable substrate* to iterate fast — modular, reproducible, fixable.
- "Affordable home robot" doesn't exist anywhere in the field. Building one (or something close to it) is a credible 1–2 year research arc and a startup-relevant skill.
- Mobile-base + arm-v2 is the cleanest path to the "in the wild" experiments already on the PARA roadmap, which the paper hints at as future work.
- Onshape (or Fusion) is a multi-year skill investment with high ROI. Now is the right time to learn it.

---

## Timeline (10–12 weeks from paper close to functional v2)

| Phase | Dates (rough) | Focus | Deliverable |
|---|---|---|---|
| **0 — Recover + paper closeout** | Jun 10 – Jun 20 | Return from TRI, camera-ready paper, half-day ArUco-board recalibration on v1 | Paper submitted; v1 arm with tighter calibration |
| **1 — Onshape skill-up + design** | Jun 20 – Jul 4 | Onshape tutorials (10–15 hrs evenings) + v2 concept sketches + parametric joint module | First Onshape arm design (rough, not manufacturable) |
| **2 — Manufacture + assembly** | Jul 4 – Jul 25 | 3D print structure, order COTS components, assemble | Mechanically complete arm, basic motion control |
| **3 — Calibration + software stack** | Jul 25 – Aug 15 | ArUco calibration with bigger boards, hand-eye for wrist cam, IK + deploy script port | Calibrated arm running existing PARA deploy stack |
| **4 — Validation + demos** | Aug 15 – Aug 31 | Re-run cup task as smoke test, fix weak points, polish | Cup-task working, v2 ≥ v1 on validation metrics |
| **5 — Mobile base + next research project** | Sep+ | Mobile base integration, then onto new research | Wheeled arm-v2, ready for "in the wild" |

Total: paper-close → arm-v2 functional in ~10 weeks. Mobile base another 4–6 weeks after that.

---

## Design parameters (preliminary — refine after TRI)

| Parameter | Target | Notes |
|---|---|---|
| **DOF** | 6 + 1 gripper | Matches custom arm v1; sufficient for table tasks |
| **Reach** | 600–800 mm | Covers typical workspace; bigger than v1 |
| **Payload** | 1–2 kg | Cup, mug, light tools |
| **Repeatability** | < 2 mm | Matches PARA's calibration floor |
| **Joint actuators** | Dynamixel XM/XH-class (or similar) | Known software stack, swappable from v1 |
| **Transmission** | Belt reduction or harmonic | Reduce backlash relative to v1 |
| **Mount** | Tabletop, with future mobile-base bolt pattern | Pre-plan for mobile integration |
| **Cabling** | Internal routing through joints | v1 lesson: external cables snag during demos |
| **Camera mounts** | Standardized workspace + wrist mount with ArUco-calibratable fixtures | Built-in calibration reference points |
| **Gripper** | Parallel two-finger with adjustable tip mount | Iteration-friendly for UMI / sticker / custom tip experiments |
| **Materials budget** | ~$2–3k | Dynamixels + bearings + 3D-print stock + electronics |

---

## TRI lessons to actively fold in

While at TRI, deliberately observe and take notes on:

1. **YAM joint design** — gear ratios, motor sizing, transmission choices. What did TRI choose and why?
2. **Wrist-camera mount approach** — physical mounting, cable routing, hand-eye calibration workflow
3. **Calibration workflow** — best practices for hand-eye, intrinsics, environment setup
4. **Error recovery + safety mechanics** — what they do when the arm hits something
5. **Workspace ergonomics** — table heights, lighting, where cameras go
6. **Bimanual considerations** — if YAM-bimanual is part of the trip, what's the dual-arm coordination story

Carry a small notebook for these. They are *the* reason to delay starting v2 until after TRI — the trip will materially improve the design.

---

## Risks + mitigations

| Risk | Likelihood | Mitigation |
|---|---|---|
| Onshape learning curve longer than budgeted | Medium | Start tutorials in evenings during Phase 0; aim for 10 hrs before Phase 1 begins |
| COTS components arrive late | Medium | Order in Phase 1, not Phase 2 — parallel-track shipping with design |
| Calibration tighter than expected requires v2.1 | Low | Budget 1 buffer week in Phase 3 |
| Scope creep into mobile base / full home robot | **High** | Mobile base is Phase 5, explicitly after v2 validation. *Do not bundle.* |
| Distraction from follow-on research papers | Medium | Treat v2 as infrastructure, not a research contribution; cap hands-on hours at 30/week |
| Burnout from too much engineering | Medium | Mix in 1–2 days/week of reading + research idea sketching during Phase 2–3 |

---

## What "done" looks like for v2

- Cup task runs on v2 at ≥ 90% in-distribution success (matches v1)
- Calibration RMSE < 0.5 px on bigger ArUco boards
- Wrist camera mountable + calibratable in < 30 minutes (vs the multi-hour v1 process)
- Modular gripper swap < 5 minutes
- Onshape source files version-controlled and parametrically modifiable (change reach → re-derive joint lengths)
- Mobile-base bolt pattern designed in but not yet integrated

When all six are true → v2 is functional and the next research project can start.

---

## Anti-goals (what v2 is explicitly NOT)

- Not a humanoid
- Not bimanual
- Not a torque-controlled high-bandwidth arm
- Not a from-scratch electronics design (use Dynamixels + standard USB cameras)
- Not a research contribution in itself — it is *substrate* for future research

If any of these creep into v2, that's mission expansion and should be a *v3 / follow-on platform*, not a v2 amendment.

---

## What this looks like a year from now

- v2 functional by end of August
- Mobile base integrated by mid-October
- "In the wild" follow-on paper experiments underway by November
- One follow-on paper drafted by early 2027
- v2 hardware design open-sourced as part of that paper
- Skill stack (CAD, manufacturing, mobile robotics) now real and resume-able for any startup pivot
- Optional: PARA-v2 paper that uses the new platform for tasks v1 couldn't reach (longer-horizon, more dexterous, "in the wild")

---

## Source documents to update when v2 begins

- `CLAUDE.md` — hardware section
- `TODO.md` — add Phase 1/2/3 tasks; mark v1 arm as "legacy, kept for comparison"
- `EXPERIMENTS.md` — note that all results pre-v2 are on v1 hardware
- A new `notes/arm_v2_log.md` for design decisions, errata, lessons-learned as the build progresses

---

## First concrete action (not now — late June)

Watch 2 hours of Onshape "intro to parametric modeling" tutorials. That's it. The skill ladder starts there. Everything else builds on it.

Until then: **this document exists, v2 is on the calendar, the urge to start now is acknowledged but deferred.**