Doc 1 · How We Got Here — Gimbal Research, Round 1 (Archive)¶

Engineered Lighting prototype series · July 2026

The project, in one paragraph¶

We're building the robotic spotlight for the Engineered Lighting fixture: a silent pan/tilt head (smart CAN servo motors, absolute encoders, no homing dance) carrying a high-CRI 3-up LED spot, living alongside the fixture's tunable-white ambient zones, exposed to Home Assistant as ordinary entities and aimed autonomously by a camera-driven perception stack — following people, lighting task surfaces and books, pointing at art when idle, never sweeping across eyes. This seven-document series takes it from research through a working bench prototype to the system architecture, with everything adversarially reviewed and every purchase specified.

The document map¶

#	Doc	What it is	You buy
1	How We Got Here	Round-1 research digest: the physics, vocabulary, and five lessons under every later decision. Archive — read for understanding	nothing
2	Choosing the Motors	The decision journey to the RMD-L-4005 smart CAN actuators — what was evaluated, why integrated actuators won	(decision feeds Doc 3)
3	Build the Gimbal	10-stage bench build: motors answering on CAN by stage 4, a balanced aimed head by stage 8. Full BoM, wiring, code, AI-partner workflow	~$185–290
4	Build the Full Fixture Bench	One ESP32-C6 running the whole fixture: 21 tunable-white channels + CC spotlight + gimbal, in ESPHome/Home Assistant	~$170–240
5	Teach It to Aim	The perception stack: cameras, ground-plane tracking, beam self-calibration, coordination, safety, user stories, phased roadmap (Phase 0 = a weekend)	~$120–320 per room
6	The Message Contract	The one page every component obeys: topics, schemas, units, watchdogs, and the dual-control architecture (HA entities + autonomy without fights). When docs disagree, Doc 6 wins	nothing
7	Building the Software	The code: pinned stack, repo layout, hardware-free testing (replay cameras, simulated fixtures), deployment, firmware growth path, licensing gates	nothing

Reading paths: Building this weekend? → Doc 3, then 4 (skim their concepts sections; Docs 1–2 optional background). Understanding the choices? → 1 → 2, then skim 5. Writing the software? → 6 → 7, with 5 as the spec. Total prototype budget, all hardware: roughly $475–850 for one room end-to-end.

Shopping list for this document: nothing¶

This is background reading — all purchases live in Docs 3–5. What this doc carries is the understanding: the physics, the vocabulary, and five lessons every later decision rests on.

The question we asked¶

Can the fixture's fixed spotlight become robotic — a motorized pan/tilt head that aims light anywhere in the room, silently, controlled by software? Round 1 surveyed how everyone else has ever built "a light that moves": stage lighting, camera gimbals, hobbyist builds.

Concepts (plain English — these terms appear in every later doc)¶

Moving head / yoke: the standard mechanical layout for a robotic light. A base rotates left–right (pan); a U-shaped bracket (the yoke) rides on it and tilts the lamp up–down (tilt). Every DJ club light is built this way.
Torque: twisting force. Motors are rated by how much they can exert (we use newton-centimeters, N·cm). The tilt motor must fight gravity trying to flop the lamp head downward.
Stepper motor: moves in fixed clicks ("steps"), no feedback — you count steps to know position. Cheap, precise, everywhere in 3D printers.
Microstepping / silent drivers: electronics that slide a stepper between its clicks in tiny increments. This is what makes modern 3D printers quiet — a measured drop from 74 dB to 35 dB just by swapping the driver chip (TMC-series "StealthChop").
Hobby servo: a small motor+gearbox+position sensor in one box (the RC-car part). Convenient, but the cheap internal sensor "hunts" — constantly twitching, audibly, when holding a load.
BLDC gimbal motor: the pancake-shaped brushless motors inside camera stabilizers. Direct drive (no gears at all), which is why camera gimbals are silent and smooth.
Encoder: a position sensor on a joint. An absolute encoder knows the angle the moment power comes on — no "finding home" ritual.
Homing: how a motor system without absolute encoders finds a reference point — usually by driving into a switch or a hard stop at startup. Commercial moving heads do a loud full-range "homing dance" on every power-up.
Slip ring vs. service loop: two ways to get wires across a rotating joint. A slip ring (sliding contacts) allows endless rotation but adds noise and wear; a service loop (a slack coil of wire) limits rotation to a turn or so but is silent, free, and reliable.

The five lessons that survived¶

Lesson 1 — Stage lighting is the right idea at the wrong scale. Commercial "mini" moving heads weigh 1.4–3.6 kg — 10–70× our ~50–150 g spotlight head — and their fans are the loudest thing in the room (a verified purchaser on a well-liked Chauvet: "the fans on these are really loud... very noticeable when [music is] not [playing]"). Even quiet-rated theatrical fixtures measure 39.6–49.5 dBA — too loud for a living room, where the ambient floor is ~30–40 dBA. Design target set here: no fans, motion ≤~35 dBA, zero noise parked.

Lesson 2 — The torque problem is small, and balance nearly erases it. Worst case (150 g head hanging 6 cm off the tilt axis): torque = mass × gravity × radius = 8.83 N·cm. But balance the head so its center of mass sits on the tilt axis (like every camera gimbal requires) and the requirement collapses to ~1.5 N·cm — small enough for thumb-sized motors, and it means near-zero current (silent, cool) while holding position for hours. Balance became a design requirement, not a nicety.

Lesson 3 — Silence favors gearless designs and joint encoders. The quietest documented builds either use steppers on silent drivers (good) or direct-drive BLDC with closed-loop control (best — a documented robot head on gimbal motors: "nearly completely silent," ~350 mA worst-case hold). Hobby servos lose on hunting noise; worm gears self-lock nicely (motor fully off while holding!) but waste >50% of torque to friction and can creep under building vibration. And putting the absolute encoder on the joint itself means the fixture wakes up already knowing where the beam points — the loud homing dance is deleted from the product.

Lesson 4 — Skip the slip ring. Commercial movers all chose limited rotation (540° pan) with a wire bundle over slip rings — and our beam power at 24 V is under 1 A, easy for a service loop of flexible silicone wire. Continuous rotation is a capability the coordinator doesn't need.

Lesson 5 — There's no shortcut through cheap DMX moving heads. Driving a $100 club light over DMX as a prototype fails three ways: 8-bit pan resolution (visible ~2° jumps on slow sweeps), loud fans, and the power-up homing dance. Buy one only as a teardown donor, never as the product path.

What this round got wrong (and round 2 fixed)¶

Round 1's recommendation — NEMA-11 stepper motors + belt reduction + silent driver chips — was reasonable but built everything by hand: motor mounts, belts, tensioners, encoder mounting, motor-control firmware. Round 2 found the same outcome available as a sealed smart actuator (motor + encoder + tuned controller in one purchasable part, commanded digitally) — fewer parts, fewer failure modes, and it deleted most of round 1's engineering. Read Doc 2 for that decision.