Choosing the Motor
Choosing the Motor: Torque Decides, Not RPM
Picking a motor for a robot joint or a drive? Torque decides, not RPM. A gearbox trades speed for torque, so size from the load first, then choose between stepper, BLDC with FOC, servo and brushed DC, joint by joint.
Pick a motor the way most teams pick one and you will buy the fastest motor in the catalogue, bolt it to the joint, and then discover that it cannot lift the load. The motor spins beautifully on the bench. It stalls the first time the arm reaches out horizontally.
That is because the question is not how fast the motor spins. The question is how hard it twists. A motor's job is to twist against a load, to hold a weight up against gravity, to push a wheel over a bump, to accelerate a mass from still to moving. Speed is what is left over after the twisting is done. A sprinter who cannot lift the weight does not help you move the weight.
This article is the sizing method. It starts with the load, works out the torque the joint actually needs, runs that number through the gearbox, and only then asks which of the four motor families, stepper, brushless DC with field-oriented control, servo, brushed DC, fits each joint. The answer is almost never the same motor at every joint. A robot arm is not one motor decision. It is one decision per joint, and the shoulder and the wrist want different answers.
The words, in plain English
Torque is twisting force. It is measured in newton-metres (N·m). One newton-metre is the twist you get from pushing with one newton, about the weight of a small apple, at the end of a one-metre lever. Hold a 1 kg bag of sugar at arm's length and your shoulder is supplying torque. The longer the arm, the more torque the same bag needs.
RPM is revolutions per minute, how fast the shaft spins. A motor that spins at 10,000 RPM is fast. It is not necessarily strong. Speed and strength are different things, and a gearbox trades one for the other.
A gearbox (also called a reducer) is a set of gears between the motor and the load. It slows the motor down and multiplies its torque by the same ratio. A 100:1 gearbox turns 100 motor revolutions into one output revolution, and multiplies the motor's torque by roughly 100, minus what the gears lose to friction.
Reduction ratio (N) is the gearbox's number. N = 100 means the motor spins 100 times for every one turn of the output shaft.
Efficiency (η) is the fraction of the power that survives. A motor that is 90% efficient turns 90% of the electricity into motion and 10% into heat. A gearbox that is 93% efficient loses 7% of the torque to friction.
Holding torque is the torque a powered motor resists with when it is stopped. It is the big number on a stepper motor's datasheet, and, as this article will show, the number most likely to mislead you.
Pull-out torque is the torque a stepper can actually deliver at a given speed without losing steps. It is the number you should size by. It falls as speed rises.
Continuous torque is the torque a motor can deliver all day without overheating. Peak torque is what it can deliver for a few seconds. Size on continuous. Never size on peak, and never size on stall torque, the usable continuous torque is roughly half of a motor's stall torque.
Backdrivability is whether you can turn the motor by pushing on the output shaft. A backdrivable joint can be pushed by hand, which is what lets a collaborative robot feel a person touching it. A joint behind a high-ratio gearbox usually cannot be backdriven, the gears lock it in place.
FOC (field-oriented control) is the control method that makes a brushless motor behave. It measures the motor's three phase currents, transforms them into a rotating frame of reference, and controls the torque-producing current directly. Think of it as steering the magnetic field rather than switching the coils on and off.
An encoder is a sensor on the motor shaft that reports exactly where the shaft is. A motor with an encoder knows its position. A motor without one is guessing.
Open loop means the controller sends commands and never checks the result. Closed loop means the controller measures what actually happened and corrects the error. A stepper is usually open loop. A servo is closed loop by definition.
Cogging is the small jerky pull you feel when you turn a brushless motor's shaft by hand, the magnets snapping from one pole to the next. It makes the motor less smooth at very low speed.
Backlash is the dead movement in a gearbox: the output shaft can wiggle a little without the motor moving, because the gear teeth have clearance. It is measured in arcminutes. One arcminute of backlash on a one-metre arm moves the tool by about 0.29 mm.
Thermal derating is what happens when a motor gets hot: it has to deliver less torque, or it cooks itself. A motor sized with no margin derates itself the first warm afternoon.
The rule: torque decides, not RPM
A gearbox trades speed for torque at ratio N. That is the whole rule, and it is why RPM is the wrong place to start. The motor's speed is whatever the gearbox leaves over. The torque is the thing you cannot get back.
The two equations that matter are these:
- τ_motor = τ_load ÷ (N × η_g), the motor torque you need is the load torque divided by the gear ratio and the gearbox efficiency.
- ω_motor = ω_load × N, the motor speed is the load speed times the gear ratio.
Read them together and the design process writes itself. First, work out the torque the joint needs. Second, pick a gear ratio that brings the motor's speed into the range where that motor family is strong. Third, check the motor can deliver the torque at that speed. If it cannot, change the ratio or change the motor, do not buy a bigger motor and hope.
A worked example from a motor guide makes the point: a wheel needs 1.86 N·m (already including a 1.5× safety factor). The wheel turns at 119 RPM. With a 20:1 gearbox at 85% efficiency, the motor needs 1.86 ÷ (20 × 0.85) = 0.109 N·m, and it spins at 119 × 20 = 2,380 RPM. That is a small motor doing a big job, because the gearbox did the multiplying. The gearbox, not the copper, supplies the torque.
The same guide states the rule in one line: torque decides the motor, not RPM. Size the motor from load torque first.
The four torque terms, and the margin
A real joint needs more than the textbook T = F × r. Four terms add up to the torque the motor must deliver, and then the safety factor multiplies the total:
Static torque is holding the load still against gravity. For an arm held horizontally, T = m × g × r, where m is the mass beyond the joint, g is 9.81 m/s², and r is the distance to the centre of mass. It is usually the largest term, and it is at its worst when the arm is horizontal. Point the arm straight up or straight down and the gravity torque is roughly zero, the load's line of action passes through the joint.
Dynamic torque is accelerating the load: T = J × α, where J is the moment of inertia and α is the angular acceleration. A heavy load far from the joint has a large moment of inertia, so it needs a large torque just to start moving.
Friction torque is the drag in the bearings, the gearbox and the seals. It is small on a clean bench and large on a dirty one.
Loss torque is what the gearbox eats. It is already inside the η_g in the gearbox equation, but it belongs in the sum when you are working out what the joint needs before the gearbox.
The safety factor is 1.5 to 2. Every guide that gives a number gives the same one: size for a 1.5×–2× margin on continuous torque. The margin is what keeps the motor out of thermal derating, absorbs the friction you forgot, and covers the day the payload is 20% heavier than the specification said. An open-loop stepper that runs out of torque stalls or skips steps, and the controller does not know it has happened, so for a stepper, the margin is not optional. It is the only thing standing between the arm and a silent failure.
And size the joint from the full load, not the payload alone. A joint never supports only the customer's payload. Every link beyond that joint has mass, and each link's weight acts through its own centre of mass. Ignoring link mass badly underestimates torque, especially in long arms. The shoulder carries the weight of everything beyond it; the elbow carries the forearm; the wrist carries only the tool and the payload. That is why the joints near the base are physically larger and use higher-ratio transmissions.
The four families
The stepper: the honest motor that lies about its strength
A stepper is a brushless motor with a large number of magnetic poles, built so that instead of spinning freely it snaps to a sequence of discrete positions, steps. A standard hybrid stepper moves 1.8° per full step, 200 steps per revolution (50 rotor teeth × 4 steps per tooth pitch). A 0.9° version has 100 rotor teeth and gives 400 steps per revolution.
The controller positions it by counting pulses, with no feedback from the motor. That is the stepper's superpower: open-loop positioning with zero tuning and zero feedback hardware.
It is also its trap. The headline number on the datasheet is the holding torque, the torque a powered motor resists with when it is stopped. But a stepper's torque falls as its speed rises, because the winding's inductance limits how quickly the current can build up on each step, and at higher step rates there is less time for it to do so.
A NEMA 23 stepper rated at 2 N·m holding torque may deliver only 0.5 N·m at 1,000 RPM, a 75% loss. The single most expensive mistake in stepper selection is sizing on holding torque and ignoring the torque-speed curve. A stepper picked on holding torque alone will stall the first time it accelerates a real load.
Size on pull-out torque at your operating speed.
The stepper's numbers, checked against the guides: high torque at low speed, in the 0–500 RPM range, with torque mostly gone above 1,000 RPM. Open-loop positioning accuracy around ±0.05° unloaded, and note that microstepping improves resolution, not accuracy: a 1.8° stepper is accurate to maybe ±5% of a full step no matter how finely you slice it, and most microsteps carry so little incremental torque they cannot move the load against friction.
A 1/16 microstepping driver reaches 0.056° nominal resolution, but the real accuracy is limited by detent torque and the load. Efficiency 40–70%, because a stepper draws its full rated current even when it is standing still, to maintain holding torque, and all of that current becomes heat. Cost $20–80 for the motor.
The rule the guides give is simple: if your axis spends most of its life holding a static position at low speed, and you can afford a 2× torque margin, a stepper is almost always the cheapest correct answer. 3D printers, light arms, lab automation stages, valve actuators, that is the stepper's home.
The BLDC with FOC: the efficient motor that needs a brain
A BLDC motor is a brushless DC motor: a permanent-magnet rotor, stationary windings, and electronic commutation instead of brushes. With field-oriented control, the drive measures the three phase currents, converts them into a rotating reference frame (the Clarke and Park transforms), and controls the torque-producing current directly, torque is proportional to Iq, the quadrature current, and Id is regulated to zero. Two PI loops, one per axis, run at 10–40 kHz, which gives smooth torque down to near-zero speed.
The numbers, checked: efficiency 85–95%, the guides agree on 85–92% for a typical system and up to 95% with FOC. Best torque density of the four families: a standard BLDC delivers roughly 0.5–1.5 N·m/kg, a high-torque design 2.0–4.0 N·m/kg. Speed range 2,000–10,000 RPM in the typical rated range, with high-speed inner-rotor designs reaching 20,000 RPM.
Position accuracy is encoder-limited, better than 0.01° with a good encoder. And it is fast: a Moog BLDC simulated in PSIM reached its nominal 2,837 RPM in 0.02 seconds, and a tuned drive settled in about 0.025 seconds; mechanical time constants run 2–50 ms depending on the motor's size. That speed is load-dependent, the figure belongs to a small motor, not to a loaded arm, but the order of magnitude is right, and it is an order of magnitude faster than a stepper can accelerate a load.
The property that matters most for a battery-powered machine is how the current behaves. A BLDC draws current proportional to the actual load. At 50% load it draws roughly 50% of its rated current.
At idle, with no load, the current drops to near zero. A stepper draws its rated current continuously, regardless of load. For the same 500 W of mechanical output, a BLDC at 90% efficiency consumes 555 W from the supply; a stepper at 60% efficiency consumes 833 W, 50% more power, all of which becomes heat.
And a BLDC lasts: 20,000+ hours of service life against 1,000–3,000 hours for a brushed motor, because there are no brushes to wear out.
The rule: choose BLDC with FOC when efficiency, speed, or backdrivability matters, drones, legged robots, AMRs, and any joint that has to feel what it is touching. A BLDC with FOC is backdrivable, which is what lets a collaborative robot sense force through its motors. That is why the force-sensing joints in Article #10's arm want BLDC, not steppers.
The servo: the package, not the motor
Here is the fact that separates a servo from the other three: a servo is not a motor type. It is a motor plus feedback plus a correcting drive, sold as one package. An AC servo is essentially a brushless motor with a high-resolution encoder and a servo drive. A DC servo is a brushed or brushless motor with the same treatment. The motor technology overlaps heavily with BLDC. The difference is the loop around it.
A real servo stacks three nested loops, each feeding the next: a current (torque) loop, the fastest, which sets the motor's torque; a velocity loop around it; and a position loop on the outside. The position loop says where the shaft should be, the velocity loop says how fast it should get there, and the current loop says how hard to push. Tuning those three gains is the art of servo commissioning, and it is the same PID thinking as Article #3, with one loop wrapped around another.
The numbers, checked: a smart servo (the Dynamixel class) positions to ±0.088°; a servo with a high-resolution encoder is sub-degree, down to 0.01° or better. The encoder runs 2,500 to 1,000,000+ pulses per revolution. Speed control accuracy is ±0.01–0.1% with an encoder, against ±2–5% for a Hall-sensed BLDC. Cost is the price: a servo system typically costs 2–5× an equivalent BLDC system for the same power rating, $300–800 for a 500 W system against $80–200 for the BLDC motor and driver.
The rule: choose a servo when precision and dynamic performance under a varying load matter, CNC, pick-and-place, a joint whose load changes every cycle. You are paying for the package: the motor, the encoder, the drive, and the three loops already tuned to work together. If you can build that package yourself from a BLDC and an encoder, you can have most of the performance for a fifth of the price. If you cannot, the servo is the honest buy.
The brushed DC motor
A brushed DC motor is the oldest of the four: a wound rotor, a mechanical commutator, and brushes that switch the current as the shaft turns. It is simple, it is cheap, and it works. Efficiency 60–75%. The brushes wear out, 1,000–3,000 hours of service life against 20,000+ for a BLDC, and they make electrical noise and dust, which is why they are unwelcome near sensitive electronics.
The rule: use a brushed DC motor when cost dominates and nothing else does, a cheap wheeled base, a toy, a fan that runs at one speed. The moment the job needs efficiency, precision, or a long life, the brushed motor stops being the cheap answer, because you pay for it in batteries, in heat, and in replacements.
The gearbox is half the decision
The motor is half the actuator. The gearbox is the other half, and for a robot joint it is often the more expensive half. Three architectures matter:
Planetary gears are the workhorse: a sun gear drives planet gears inside a fixed ring. 95–97% efficiency per stage (92–94% for a two-stage box, which is what a 100:1 ratio needs), ratios of 3:1–10:1 per stage and up to 100:1 in two stages, input speeds of 5,000–10,000 RPM, and backdrivable, you can turn the output by hand. The cost is backlash: 3–15 arcminutes for a precision grade, down to 1–3 arcminutes for the best grade. One arcminute of backlash on a one-metre arm moves the tool by about 0.29 mm, so the backlash budget and the positioning budget are the same number.
Harmonic drives (strain wave gears) are the precision choice: a wave generator deforms a flexible spline inside a rigid ring, giving 50:1–160:1 in a single thin stage with essentially zero backlash, under 1 arcminute, and up to 3× the torque density of a planetary in the same frame. The price is efficiency: 65–85%, typically around 80% at rated load, and worse at light load, one catalogue example works out at 70% for a drive running at less than half its rated torque. And they are not backdrivable at high ratios, and they are limited to 3,000–4,000 RPM input because the flexspline fatigues faster at high speed. A planetary of the same frame typically costs 30–50% less than a harmonic drive.
Cycloidal (RV) drives are the heavy-duty choice: 80–90% efficiency, ratios of 30:1–190:1, and they survive shock loads over 500% of rated torque. They belong on industrial robot bases and shoulders, not on a bench cobot.
The pattern the industry has converged on is a split: harmonic drives for upper-body precision joints, where zero backlash matters; two-stage planetary gearboxes for lower-body power joints, where efficiency and cost matter. The demo below follows the same split.
The loops, from the outside in
Article #3 built a PID loop. The motor choice decides how many loops you actually have, and what each one costs you.
A stepper has no loop. The controller counts pulses and assumes the motor followed them. A missed step is silent, the controller does not know, the position is wrong, and nothing in the system will tell you. That is why the stepper's safety margin is not optional, and why the guides say to add an encoder and stop pretending it is open-loop if the load can ever exceed the torque.
A BLDC with FOC has two loops: the Id and Iq current loops, running at 10–40 kHz. They control torque directly, torque is proportional to Iq, and they give smooth torque down to near-zero speed. Add an encoder and a position loop on top, and you have built a servo out of parts. That is the configuration the cobot's force-sensing joints need: the current loop is what lets the arm feel how hard it is pushing, because current and torque are the same number once you know the torque constant.
A servo has three loops, already stacked and already tuned: current inside velocity inside position. You do not build them. You tune their gains. The position loop is the outermost and the slowest; the current loop is the innermost and the fastest, clocked at roughly 10 kHz so that the torque can correct position or speed as quickly as the mechanics can react.
The point for the buyer is not which loop is cleverest. It is which loop you are willing to own. A stepper asks you to own nothing and gives you no warning when it fails. A BLDC with FOC asks you to own the current loops and the encoder. A servo asks you to own the tuning. Pick the loop you can actually maintain.
The battery
Example Labs is a made-up company we use for worked examples.
Article #6 built the power budget for the Example Labs field sensor hub, a CR2032 coin cell, 230 mAh, 2,484 joules, and a radio that costs 30 microjoules a packet on BLE. A motor does not belong in that budget. A motor's current draw is the budget's biggest line, and it is big enough to make the radio's cost disappear.
The numbers are in the demo, but the principle is short. A motor's electrical power is its mechanical power divided by its efficiency, and the efficiency gap between the families is the battery gap. At 500 W of mechanical output, a BLDC at 90% draws 556 W and a stepper at 60% draws 833 W.
On a 48 V, 100 Ah pack, 4,800 Wh, that is 8.6 hours against 5.8 hours, for the same work. And the stepper's worst habit is not its efficiency under load. It is its current at standstill: a stepper holding a joint draws its full rated current while doing no work at all.
Two phases at 2 A on a 24 V supply is roughly 96 W, and over an eight-hour shift that is 768 Wh, 16% of the whole pack, spent holding still. A BLDC at idle draws near zero.
That is the battery argument for BLDC in one paragraph. It is also the thermal argument: the 277 W difference between the two motors at 500 W of output is heat, and heat has to go somewhere. In an enclosure, Article #7's subject, it goes into the air inside the box, and the box gets hot, and the motor derates, and the torque falls. The motor choice and the enclosure design are the same decision, made twice.
sizing the Example Labs bench arm, joint by joint
The bench is the UR5e-class cobot from Article #10: 5 kg payload, 850 mm reach, 20.7 kg total, ±0.03 mm repeatability, a 4 N force-sensor accuracy. The arm has two links of 0.425 m each. Link 1 (the upper arm) weighs 4 kg. Link 2 (the forearm and gripper) weighs 3 kg. The payload is 5 kg at the full 0.85 m reach. The demo sizes the shoulder, the elbow and the wrist, and then checks the battery and the bill.
Step 1, the load torque, worst case first. The worst case is the arm fully extended horizontally, where every weight acts at its full horizontal distance from the joint. The formula is T = m × g × r, summed over every mass beyond the joint, the arm's own weight included, because a joint never supports only the payload.
- Shoulder: link 1 at its centre of mass (0.2125 m) gives 4 × 9.81 × 0.2125 = 8.34 N·m; link 2 at 0.6375 m gives 3 × 9.81 × 0.6375 = 18.76 N·m; the payload at 0.85 m gives 5 × 9.81 × 0.85 = 41.69 N·m. Total static torque: 68.8 N·m.
- Elbow: link 2 at 0.2125 m gives 6.25 N·m; the payload at 0.425 m gives 20.85 N·m. Total: 27.1 N·m.
- Wrist: the payload at 0.15 m from the wrist joint gives 5 × 9.81 × 0.15 = 7.36 N·m.
The dynamic term at the shoulder: the moment of inertia is 5 × 0.85² + 3 × 0.6375² + 4 × 0.2125² = 5.01 kg·m². Accelerating to 1.57 rad/s (90° in one second) in half a second needs an angular acceleration of 3.14 rad/s², so the dynamic torque is 5.01 × 3.14 = 15.7 N·m, about 23% of the static torque, and the term most teams forget.
Applying the 1.5× safety factor to the total: the shoulder needs (68.8 + 15.7) × 1.5 = 126.8 N·m; the elbow needs 27.1 × 1.5 = 40.7 N·m; the wrist needs 7.36 × 1.5 = 11.0 N·m. For scale, the pipeline's worked example, 1 kg at the end of a 0.5 m arm, needs 1 × 9.81 × 0.5 = 4.9 N·m at the shoulder before the arm's own weight. The method is the same; only the numbers change.
Step 2, through the gearbox to the motor. The shoulder needs 126.8 N·m at the joint. With a 100:1 two-stage planetary gearbox at 93% efficiency, the motor needs 126.8 ÷ (100 × 0.93) = 1.36 N·m, spinning at 15 RPM × 100 = 1,500 RPM. With a 100:1 harmonic drive at 80% efficiency, the motor needs 126.8 ÷ (100 × 0.80) = 1.58 N·m, the harmonic's friction costs 16% more motor. A NEMA 23 stepper rated at 2 N·m holding torque delivers about 0.5 N·m at 1,000 RPM, and less at 1,500 RPM. It cannot deliver 1.36 N·m at 1,500 RPM. The stepper fails the shoulder, on torque at speed, not on holding torque, which is the number on the datasheet. A BLDC at 1.36 N·m and 1,500 RPM is inside its flat-torque range. It passes.
The elbow needs 40.7 N·m. With a 50:1 planetary at 93%, the motor needs 0.87 N·m at 750 RPM, past the stepper's strong range, borderline on its torque curve, comfortable for a BLDC. The wrist needs 11.0 N·m. With a 10:1 single-stage planetary at 96%, the motor needs 1.15 N·m at 300 RPM, inside the stepper's strong 0–500 RPM range, where a 2 N·m NEMA 23 delivers roughly 1.8 N·m. The stepper passes the wrist. The same arm, three joints, two different answers.
Step 3, the battery. The shoulder's mechanical power is 126.8 N·m × 1.57 rad/s = 199 W. At 90% efficiency the BLDC draws 221 W; at 60% a stepper would draw 332 W. For the 500 W comparison: a BLDC draws 556 W, a stepper 833 W, 50% more, all heat. Over an eight-hour shift that is 4,448 Wh against 6,664 Wh. On a 48 V, 100 Ah pack (4,800 Wh), that is 8.6 hours against 5.8 hours for the same work. And the standstill cost: a stepper holding a joint at 2 A per phase on 24 V draws roughly 96 W while doing nothing; over an eight-hour shift that is 768 Wh wasted holding still, 16% of the pack. A BLDC at idle draws near zero.
Step 4, the bill. For a 500 W system, the verified system costs are: stepper $20–80, BLDC with driver $80–200, servo $300–800. Take the midpoints, $50, $140, $550, and price four joints: all steppers $200, all BLDC $560, all servos $2,200, 3.9× the BLDC system. The mixed build the demo actually recommends, BLDC at the shoulder and elbow, steppers at the wrist and gripper, costs 2 × $140 + 2 × $50 = $380, and it is the only one of the three that works at every joint. The cheapest build that works beats the cheapest build, and it beats the most expensive build too.
Step 5, the decision, joint by joint. Shoulder: BLDC with FOC and a planetary gearbox, because the torque at speed kills the stepper and the force sensing needs backdrivability, which a harmonic drive does not give. Elbow: BLDC with FOC. Wrist: a stepper with a 10:1 planetary, because the speed is low, the position is repeatable, and the stepper is the cheapest correct answer there. Gripper: a stepper or a hobby servo, because the load is small and the position is the job. The motor is chosen per joint, from the joint's torque, speed and sensing needs, never from the catalogue's fastest motor, and never the same motor four times.
The decision test
Four questions, in order. Answer them before you open a catalogue.
- What is the torque at each joint, worst case? Arm horizontal, full payload, the arm's own weight included, static plus dynamic, times 1.5. If you do not have this number, you do not have a motor decision. You have a guess.
- What is the motor's torque at your operating speed? Read the pull-out torque curve for a stepper, or the torque-speed curve for a BLDC, at the speed the gearbox actually puts the motor at. The holding torque and the peak torque are not the answer.
- What does the joint have to feel? If the joint has to sense force, a cobot, a gripper, a leg, it has to be backdrivable, which rules out high-ratio harmonic drives and steppers behind gearboxes, and points at BLDC with FOC.
- What does the motor cost to run, not to buy? The current at standstill, the efficiency under load, and the heat the enclosure has to shed. A $50 stepper that eats 768 Wh a shift holding still is not the cheap motor.
Objections
"But the stepper's datasheet says 2 N·m." It says 2 N·m holding torque, at zero speed, with the motor stopped. Your joint does not run at zero speed. At 1,000 RPM that motor delivers about 0.5 N·m. Size on the pull-out torque at your operating speed, or size on the holding torque and stall the first time the arm accelerates a real load.
"But the BLDC needs a drive and an encoder, and the stepper just needs pulses." True, and it is the whole trade. The stepper's simplicity is real, no tuning, no feedback hardware, and it is why the stepper wins the wrist. But the simplicity has a price: no loop, no warning, and full current at standstill. The BLDC's drive is the price of knowing where the shaft is and how hard it is pushing.
"But the servo costs five times the BLDC." Two to five times, for the same power rating, $300–800 against $80–200 for a 500 W system. You are paying for the package: the motor, the encoder, the drive, and three nested loops already tuned to work together. If you can build that package from a BLDC and an encoder, you can have most of the performance for a fifth of the price. If you cannot, the servo is the honest buy, and the alternative is not a cheaper servo. It is a BLDC system you build yourself.
"But we can just buy a bigger motor." A bigger motor is a bigger current draw, a bigger battery, a bigger enclosure, and a bigger heat problem, and it does not fix a gearbox ratio that puts the motor at the wrong speed. The gearbox is what multiplies the torque. Change the ratio first. It is cheaper than the motor, and it is the part that actually does the multiplying.
Sources
- Techietory, "Robot Actuators Explained: Motors, Servos & Drive Systems"
What we used
(), the torque-decides rule; the gearbox equations τ_motor = τ_wheel ÷ (N · η_g) and ω_motor = ω_wheel · N, with the worked example (1.86 N·m at the wheel, 20:1 at 85% → 0.109 N·m at the motor, 2,380 RPM); the 1.5×–2× safety margin; the comparison table (brushed DC 60–75%, stepper 40–70%, BLDC + FOC 85–95%; stepper ±0.05° open-loop unloaded, smart servo ±0.088°, BLDC + FOC encoder-limited <0.01°); the FOC description (torque proportional to Iq, Id regulated to zero, two PI loops at 10–40 kHz); the selection rules (steppers for repeatable low-speed positioning without feedback; BLDC + FOC when efficiency, speed or backdrivability matters; servos when precision and dynamic performance under varying load matter)
- SHZL Drive, "BLDC Motor vs Stepper Motor: When to Use Each"
What we used
(), stepper speed range 100–10,000+ RPM for BLDC, 0–1,000 RPM for steppers with torque dropping above 500 RPM; NEMA 23 rated 2 N·m holding, ~0.5 N·m at 1,000 RPM (a 75% loss); BLDC 85–92% efficiency, stepper 50–70%; BLDC current proportional to load, near zero at idle; the 500 W comparison (BLDC at 90% consumes 555 W, stepper at 60% consumes 833 W)
- NBSHZL Motor, "BLDC Motor for AGV & AMR: Complete Drive Motor Selection Guide"
What we used
(), BLDC 85–92% efficiency against 65–75% brushed and 40–60% stepper; a BLDC-driven AGV runs 12–16 hours per charge against 8–11 hours for a brushed-motor AGV on a 48 V / 100 Ah (4,800 Wh) pack; wheel torque 1–5 N·m per wheel for light-duty AMRs; BLDC most efficient at 2,000–4,000 RPM, with wheel RPM of 38–255 RPM and gear reduction of 10:1–100:1
- HDB Motors, "Stepper Motor vs BLDC Motor: Which One for Industrial Automation?"
What we used
() and "BLDC vs Stepper Motor: When to Use Each in Automation" (https://www.hdbmotors.com/industry/bldc-vs-stepper-motor-when-to-use.html), stepper 1.8°, 200 steps/rev; stepper strong at 0–600 RPM, most torque lost above 1,000 RPM; BLDC 3,000–10,000 RPM at 85–92% efficiency; stepper draws full rated current even at idle, 40–70% efficiency
- Greensky Power, "How Stepper Motors Work: Pulses, Step Angle and Torque"
What we used
(), 1.8° gives 200 steps/rev, 0.9° gives 400, five-phase 0.72° gives 500; torque falls with pulse rate because winding inductance limits current build-up
- Robo2U, "Stepper Motors and Drivers: Microstepping, Torque Curves, TMC"
What we used
(), size on pull-out torque at top speed with a 1.5–2× margin; the most expensive mistake is sizing on holding torque; a 1.8° stepper is accurate to maybe ±5% of a full step no matter how finely it is microstepped; if the axis holds a static position at low speed with a 2× margin, a stepper is almost always the cheapest correct answer
- StepperDirect, "How to choose a stepper motor"
What we used
(), a safety factor of 1.5 to 2 on the torque you calculate is a common rule for open-loop steppers; the torque curve is measured with a particular driver, supply voltage and microstep setting
- Baolong Motor, "Exploring the Advancements and Applications of Brushless DC Motors"
What we used
(), BLDC efficiency 85–95% against 75–80% brushed; service life 20,000–50,000 hours against 1,000–3,000 hours for brushed; acceleration rates of 10,000–50,000 RPM per second. (Its "Power Density 2–3 W/kg vs 5–8 W/kg" table is clearly wrong and was not used, see Corrections.)
- Design News, "New Developments in Commutation and Motor Control Techniques"
What we used
(), FOC brushless motors reach up to 95% efficiency; FOC improves over trapezoidal commutation and offers a speed range advantage over sinusoidal commutation
- Smartotics, "Stepper Motors and Brushless DC Motors: Precision and Efficiency"
What we used
(), stepper cost $20–80, servo $50–500+; stepper speed limited to under 2,000 RPM, servo over 10,000 RPM; stepper efficiency lower because current is always on
- SHZL Drive, "BLDC Motor vs Servo Motor: Differences, Performance & How to Choose"
What we used
(), a servo closes three nested control loops (position, speed, current); servo systems cost 2–5× equivalent BLDC systems; system cost at 500 W: BLDC $80–200 (motor + driver), servo $300–800 (motor + servo drive); servo position accuracy sub-degree (0.01° or better)
- Greensky Power, "All About Servo Motors Basics"
What we used
(), a servo is a motor, a feedback device and a drive that continuously compare commanded position, speed or torque against the actual value; three nested loops (current/torque fastest, velocity, position outermost); size from peak torque, RMS torque, speed and the load-to-rotor inertia ratio
- Greensky Power, "High Torque BLDC Motor: Complete Technical Guide"
What we used
(), the gearbox equations (T = 9550 × P / n; T_out = T_motor × ratio × η); standard BLDC torque density 0.5–1.5 N·m/kg, high-torque 2.0–4.0 N·m/kg; a 0.26 N·m motor behind a 50:1 planetary at 92% gives 12.0 N·m at the output
- Retek Motion, "Brushless Motor for Robotics: Torque Density, Cogging, and Custom Integration"
What we used
(), robotics BLDC torque density 0.8–2.2 N·m/kg; wrist axis 0.5–1.5 N·m, elbow 3–8 N·m, shoulder 15–40 N·m; the derating factor between peak and continuous torque is typically 2.5×–3×, always specify the continuous figure, not the peak
- MDPI Energies 2026, "Review of High-Torque Electric Machines Applied in Biorobotics and Wearable Devices"
What we used
(), torque density above 35–45 N·m/kg is needed to approach biological muscle; conventional BLDC and PMSM dominate the market
- Academia.edu / Moog, "Assessment of Operating Characteristics of Brushless DC Motor"
What we used
(), a Moog BN42 BLDC simulated in PSIM accelerated to its nominal 2,837 RPM in 0.02 s under closed-loop control
- Machine Design, "Position control primer"
What we used
(), mechanical time constants of 2–50 ms depending on motor size; current controllers clocked at roughly 10 kHz; making moves in less than a few milliseconds is generally not feasible once load inertia is included
- ElectricalFlux, "How to Build a Robot Arm: Servo Sizing, Torque Math, and ESP32 Wiring"
What we used
(), the worst case is the arm fully extended horizontally; τ = Force × Distance; a 1 kg payload at 30 cm generates roughly 3.0 N·m at the base, 60 kg·cm with a 2× safety factor
- Projectech, "Calculate Torque & Choose a Robot Motor"
What we used
(), T_required = (T_static + T_dynamic + T_friction + T_loss) × safety factor; never size on stall torque, usable continuous torque is roughly half of it; the worked example (0.5 kg link at 0.2 m + 0.3 kg payload at 0.4 m → 2.16 N·m static, 3.24 N·m with 1.5×); size the joint from the full load, not the payload alone
- Firgelli, "Robot Arm Payload Calculator, Joint Torque"
What we used
(), τ_i = Σ(m_j × g × d_ij × cos θ_i); joint 1 always carries more, because every mass beyond it ends up at a longer arm
- ZhongYi Gear, "Planetary Gear vs Harmonic Drive vs Cycloidal Drive"
What we used
(), planetary 95–97% per stage, harmonic 70–85%, cycloidal 80–90%; harmonic ratios 50:1–160:1 single stage with under 1 arcminute backlash; planetary backlash 3–15 arcminutes; planetary costs 30–50% less than a harmonic drive of the same frame
- Sango Automation, "Harmonic Drive vs Planetary Gearbox for Humanoid Robots"
What we used
(), single planetary stage 95–97%, two-stage 92–94%; harmonic ~85% at rated load/speed; the industry split (harmonic for upper-body precision joints, two-stage planetary for lower-body power joints)
- Repunite, "Harmonic Drive vs RV vs Planetary"
What we used
(), planetary 90–97% per stage, harmonic 65–85%, RV 85–92%; one arcminute of backlash at a one-metre arm is roughly 0.29 mm of endpoint error