Doc 2 · Choosing the Motors — Why Smart CAN Actuators Won¶

Engineered Lighting prototype series · July 2026

The purchase this decision produces¶

The full build shopping list (wiring, supply, tools) lives in Doc 3's BoM. This doc's outcome is the part at the heart of it:

Part	Qty	Est.	What it is and why it's the pick
MyActuator RMD-L-4005-100-C	2	$60–120	A "smart servo": brushless gimbal motor, 18-bit absolute encoder, and a factory-tuned FOC controller sealed into one Ø39.6 × 23 mm, 65 g puck. Direct drive — no gears to whine or wear. Runs natively on the fixture's 12–24 V bus and is commanded over CAN in single 8-byte messages ("go to 32.5° at 10°/s"); all motion control happens inside the part. One becomes the pan axis, one the tilt. Ordering note: select the -C (CAN) variant — the same listings carry an RS485 "-R" twin, and "-25T" is this motor's deprecated old name. US sources: Amazon, RobotShop, Dings Motion USA
Caddx GM2 bare 2-axis gimbal (optional)	1	$70	A working 30 g UART-commanded FPV gimbal — a great study article for control feel and micro-scale mechanical packaging (its motors are sized for 5–20 g cameras, so it's a reference, not the actuator)

Everything else evaluated — bare motors + DIY control stacks, hobby servos, drone motors, the other integrated actuators — is covered in the decision journey below, so the reasoning stays legible without cluttering the shopping list.

Concepts (plain English)¶

BLDC (brushless motor): three coils of wire pushing a ring of magnets. No contacts to wear, no gears — silent by construction. But it's "dumb": something must energize the right coils at the right moments.
FOC (field-oriented control): the math that drives a BLDC smoothly — steering the magnetic field continuously instead of in steps. FOC done well = silky, silent motion; FOC done badly = buzzing and vibration. Tuning FOC yourself is the hard part of DIY.
Absolute encoder: a magnetic angle sensor on the shaft that knows the position instantly at power-on. The RMD-L's is 18-bit ≈ 0.001° resolution — about 170× finer than we need. One honest limit: it's single-turn — it knows the angle within one revolution but can't count full turns made while unpowered (irrelevant if the frame has hard stops, which ours will).
Integrated smart actuator: motor + encoder + FOC controller + firmware sealed in one unit, commanded over a digital bus. You skip motor drivers, encoder mounting, and control-loop tuning entirely. This category is why round 1's plan died.
CAN bus: a rugged two-wire network from the automotive world. Every device has an address; messages are 8-byte packets; wiring is one twisted pair shared by all devices. Our ESP32-C6 has a CAN controller on-chip (Espressif calls it TWAI).
KV rating: motor speed per volt. High-KV motors (drone racing) are wound to spin fast; gimbal motors are low-KV, wound for smooth torque at near-zero speed. Same size, opposite personalities — why cheap tiny drone motors can't do this job.
Cogging: the magnetic "clickiness" you feel turning a motor by hand — magnets snapping between preferred positions. Visible as tiny stutter at very low speeds; good gimbal motors and good FOC minimize it.
Direct drive: load bolted straight to the motor, no gearbox. Zero backlash, zero gear noise. The whole camera-gimbal industry is direct drive — and so are we.

The decision journey (what we searched, in order)¶

1. "Use what DJI uses" — a wall. The Osmo-Pocket-scale motors (~Ø12–15 mm) are proprietary: no published dimensions, no purchasable parts, no pinouts. Teardowns show what's inside (a standard driver chip, MP6536, plus Hall sensors and flex cables — no slip rings), but the connectors are undocumented and the newer devices speak private digital buses. Even controlling a DJI phone gimbal over Bluetooth is a dead end — the best reverse-engineering effort gets telemetry but the device silently ignores motor commands. Verdict: don't build on DJI hardware.

2. The buyable floor. The smallest true gimbal motor money can buy is the CopterLab PM1105 — Ø12×16 mm bare, 12 g, ~1 N·cm, $21.71. Genuinely Osmo-scale, but right at our torque floor (needs a very strictly balanced head). The workhorse class is one size up: 2804-class (Ø35 mm, ~50 g, 2.5–3.4 N·cm), sold with factory-mounted encoders. Nothing between ~Ø18 mm and Ø35 mm ships with an encoder — a real market gap, confirmed by the builder community.

3. The DIY path and its documented pain. Bare gimbal motor + encoder + open-source FOC (SimpleFOC) works — there are lovely precedents (a "practically noiseless" pan/tilt camera turret; a fully open-source smartphone gimbal with CAD and PCBs). But the forums also document the failure modes honestly: encoder mounting quality is the #1 killer, low-speed cogging resists tuning, and open-loop shortcuts are never quiet. Budget: weeks. And the convenient two-axis driver board we'd have used (MKS Dual FOC) turned out to have a community reliability record bad enough to withdraw it.

4. The category that ends the debate: integrated smart actuators. Robotics vendors now sell exactly what we were about to hand-build, factory-tuned and sealed:

Actuator	Size	Torque (nom/peak)	Bus	Volts	Price/axis	Verdict
MyActuator RMD-L-4005	Ø39.6×23 mm, 65 g	7 / 25 N·cm	CAN + RS485	12–24 V	~$30–60/axis	Winner — direct drive, 18-bit absolute encoder, native on our 24 V bus, US distributor + Amazon
M5Stack RollerCAN	40 mm cube, 83 g	6.5 N·cm @16 V	CAN/RS485	6–16 V	$37–44	Cheapest; not 24 V; torque marginal
SteadyWin GIM3505 (driver SKU)	Ø43×30 mm	—	CAN/RS485	12–48 V	$76–120	Fine; China-only shipping; SKU confusion (encoder-only vs driver versions)
CubeMars GL40 II	Ø46×33.5 mm, 125 g	25 / 68 N·cm	CAN	16 V	$134	The premium "quiet gimbal actuator" if RMD disappoints
DYNAMIXEL XL330	20×34×26 mm, 18 g	10 N·cm	TTL serial	5 V	$27	Tiny and charming — but geared (plastic gears = noise) and 5 V. Wrong physics for silence

Why the RMD-L-4005 wins, in product terms: each axis becomes a part instead of a project. The fixture always knows where the beam points (absolute encoder → no homing dance, survives power loss). Torque headroom (25 N·cm peak vs 8.83 worst-case unbalanced) makes head balancing good-practice rather than mandatory. And the electronics shrink: the ESP32-C6 already has CAN on-chip, so the whole motion subsystem needs one $2 transceiver chip — no motion co-processor, no driver boards, no encoder wiring.

The trade-offs, honestly: the control loop is a black box (if it whines at hold, you can't retune it — that's the first bench measurement in Doc 3); it's ~5 mm fatter than the smallest bare motors; it's a Chinese vendor with a US distributor; and at production volume you'd cost-down to your own driver on the fixture PCB — the RMD-L is the prototype accelerant and the behavioral reference for that later design.

5. Reality checks from adjacent worlds. FPV drone gimbals (Caddx GM2: 30 g, $70, PWM/UART, even natively supported by ArduPilot autopilots) prove the control patterns but are motored for 5–20 g cameras — a payload class too small for our head. And architectural lighting already has a patented motorized recessed spotlight line (Forma MOTOLUX, ±40° dual-axis, DMX/Casambi — patent US 11215345): the category exists commercially, validating the concept and requiring a freedom-to-operate review before our commercial fixture.

Risks carried forward¶

Hold-state whine is the one unknown that matters: a statically held BLDC can whine at some operating points, and the RMD's sealed loop can't be retuned. Doc 3, stage 5 measures it before anything else depends on it.
Speed vs. silence: follow-me needs 54–80°/s pan on close passes (Doc 5's math); noise at those speeds is unmeasured — same stage-5 bench item.
Protocol drift: older stock has shipped with older protocol docs — trust the PDF in the box over any byte layout written here.
Patents: Forma (motorized recessed fixtures) and Position Imaging US 12,190,542 (beam self-calibration, see Doc 5) — counsel review before commercialization.

Doc 2 · Choosing the Motors — Why Smart CAN Actuators Won¶

The purchase this decision produces¶

Concepts (plain English)¶

The decision journey (what we searched, in order)¶

Risks carried forward¶

Further reading¶