AutomateDirectUSAMotion LibraryShop the parts catalog

Selection guide

Belt or Gearbox Reduction for a Stepper Robot Arm

Servo and stepper motion Published 8 min read

Short answer

Size each joint for worst-case gravity torque first: an 8 kg link at 300 mm reach needs 23.5 N·m, which a 2 N·m-class stepper meets only through roughly 14:1 — past what one practical GT2 pulley stage delivers, so that joint takes a gearbox. The motor-shaft magnetic encoder sees no jumped teeth or arm flex, so a home switch and travel-limit switches are needed either way.

On this page
  1. Size each joint at its worst pose
  2. Worked example: shoulder joint, 8 kg link at 300 mm reach
  3. When a timing belt meets the requirement
  4. Belt versus gearbox at a glance
  5. What a motor-shaft magnetic encoder cannot see
  6. Fit a home switch and travel limits anyway
  7. The decision, case by case
  8. Questions asked

Decide this with arithmetic, not preference. Each joint of a stepper-driven arm has a worst-case gravity torque — mass, gravity, perpendicular reach — and that number, not habit, picks the transmission: a timing belt stage is the right choice when its pulley reduction, tooth engagement and tension capacity meet the computed requirement, and a gearbox is the right choice when they do not. Separately, a closed-loop stepper's magnetic encoder closes the loop around the motor shaft only. It cannot see a jumped belt tooth, a pulley slipping on its clamp, or a flexing printed arm, so a home switch and travel-limit switches belong on the machine either way. The sections below give the sizing method, work a shoulder joint end to end, and end with the decision for each case.

Size each joint at its worst pose #

Gravity sets the floor. Put the arm in its most demanding pose — horizontal and fully extended, payload fitted — and sum the torque each mass adds about the joint being sized: τ = m·g·r_perp, with r_perp the perpendicular distance from the joint axis to that mass's centre of gravity. Include the payload, every downstream link, and each downstream motor, bracket and cable; a long reach dominates the sum even when the payload is light.

Motion adds acceleration torque from the intended trajectory plus friction, so size to the trajectory, not to a static lift.

Reduction converts motor torque into joint torque. For a stage of ratio N = motor speed ÷ joint speed, τ_joint ≈ η·N·τ_motor, where η is the stage efficiency and τ_motor is the motor's torque at the speed the ratio imposes — not the holding-torque headline, because available stepper torque falls as speed rises. More reduction multiplies torque and motor speed alike, so confirm the motor still delivers at N × joint speed: the same margin question as checking servo speed margin for a rack-and-pinion axis.

The inputs, with units and where each comes from:

  • Link mass m = 8 kg — the link mass a published robot-joint torque walkthrough uses as its example; a real arm adds payload and every downstream link to this.
  • Perpendicular reach r = 300 mm (0.300 m), arm horizontal — an assumed bench-scale reach; substitute your own worst pose.
  • Gravity g = 9.81 m/s².
  • Motor torque τ_motor = 20 kgf·cm ≈ 1.96 N·m maximum — the shoulder-motor size one published hobby build reports; treated as best case, because torque falls with speed.
  • Stage efficiency η = 0.90 — a rule-of-thumb figure for a toothed-belt stage.
  • GT2 pitch 2 mm; motor pinion 16 teeth.

The rule, in words: gravity torque about the joint axis equals mass × gravity × perpendicular distance, and the stage ratio must satisfy N ≥ τ_joint ÷ (η × τ_motor), with motor torque evaluated at the speed the ratio imposes.

The arithmetic:

  • Gravity torque: 8 kg × 9.81 m/s² × 0.300 m = 23.5 N·m.
  • Minimum ratio: 23.5 N·m ÷ (0.90 × 1.96 N·m) = 13.3, so 14:1.
  • Speed cost: a 30 rpm joint slew (assumed) runs the motor at 14 × 30 = 420 rpm, where available torque sits well below 1.96 N·m — redo the ratio with the curve value.
  • Stage size: 14 × 16 = 224 teeth on the joint pulley, pitch diameter 224 × 2 mm ÷ π = 143 mm; each extra 1:1 on a 16-tooth pinion adds 32 mm ÷ π ≈ 10 mm of joint-pulley diameter — 5:1 needs a 51 mm pulley, 14:1 needs 143 mm.
  • Engagement: 16 teeth × (120° wrap ÷ 360°) ≈ 5 teeth in mesh at a typical wrap, under the six-teeth minimum rule of thumb.
  • Tight-side tension at the joint pulley: 23.5 N·m ÷ 0.0713 m = 330 N.

The result: this shoulder needs 23.5 N·m at the joint. A 2 N·m-class stepper reaches it only through roughly 14:1, and one GT2 stage at 14:1 means a 143 mm joint pulley, about 330 N of tight-side tension, and a 16-tooth pinion below the engagement minimum.

The decision: a single GT2 pulley stage stays practical to about 5:1 in a bench-scale joint — a 51 mm pulley — and only with 18 pinion teeth or more, or an idler adding wrap, to keep six teeth engaged. This joint needs 14:1, so it takes a planetary gearbox rated at or above 23.5 N·m continuous output at 14:1 or more, with the ratings taken from that gearbox's datasheet. The all-belt route exists — published low-cost arm designs reach large single-stage reductions with a capstan, a small drive radius in place of a small pinion — or as two rigid stages with metal pulleys and reinforced belts; either way, the 330 N tension figure is what printed brackets and pulleys must survive, and builders report cheap GT2 slipping on printed pinions long before loads like that.

When a timing belt meets the requirement #

A belt carries torque through tooth engagement, so the checks are specific:

  • Ratio comes from tooth counts: N = joint-pulley teeth ÷ motor-pinion teeth. Reduction without a gearbox is legitimate, and stacking two rigid stages multiplies the ratios.
  • Profile decides capacity and backlash. GT-class teeth engage more gradually than trapezoidal shapes and hold lower backlash; published profile comparisons put GT load capacity up to 25% higher at the same pitch, and belt makers sell GT2 lines specifically for higher capacity and longer life than trapezoidal belts. The trade-offs between profiles for drive duty are compared in this pulley tooth-profile selection guide for servo drives.
  • Engagement decides whether teeth carry or skip. Keep at least six teeth in mesh on the smaller pulley — a rule of thumb — by enough pinion teeth or an idler for wrap.
  • Tension comes from the belt supplier's installation specification, not from tightening until the arm holds. Overtension loads the motor and joint bearings and bends printed brackets; a bracket that flexes under load moves the pulley centres and drops teeth out of engagement even when the standing belt feels tight.
  • Materials follow the load: metal pulleys and steel-reinforced belts on gravity-loaded joints.

When the sizing passes, the joint takes a timing belt in the profile and pitch you sized, matched to both pulleys.

Belt versus gearbox at a glance #

Criterion Timing belt stage (metal pulleys) Planetary gearbox Printed gear train
Practical ratio About 5:1 single stage in a bench envelope (16-tooth pinion, 2 mm pitch); stack stages for more Any documented ratio; take it from the datasheet Any tooth count, but accuracy decays with wear
Torque source η·N·τ_motor, motor torque taken at the imposed speed Documented rated output torque Unrated — printed teeth deform, with no datasheet to fall back on
Backlash Tooth clearance plus belt stretch, minimised by correct installation tension Documented backlash class from the datasheet Highest, and builders report it grows worse than belts
Blind spot for a motor-shaft encoder A jump moves the joint by whole teeth at once Wind-up and backlash hide smaller errors Tooth wear and shaft deflection
Power-off holding None None unless the datasheet states it None
Choose it when The computed ratio fits one or two rigid stages and tension and engagement check out Ratio, torque or stiffness exceeds what belt stages deliver Prototyping motion only, with nothing worth protecting downstream

What a motor-shaft magnetic encoder cannot see #

A closed-loop stepper corrects only what its sensor sees, and an AS5600-class magnetic chip sits on the motor shaft. Everything downstream of that shaft is invisible to the loop: a belt tooth jumping under load, a pulley creeping on its clamp screw, gearbox wind-up, a printed arm flexing at the shoulder. The drive reports in position while the joint sits somewhere else — no fault, no alarm.

Whether position survives a power cycle is a question for the motor-and-drive documentation, not an assumption: magnetic chips read angle absolutely within one revolution and do not count revolutions while unpowered, so treat every power-up as needing a fresh reference.

One safety note: if a joint can fall under gravity when power drops or the drive disables, design the restraint in — a brake or counterbalance, because neither a belt stage nor a standard planetary holds by itself — and size the stopping arrangement through the machine's risk assessment; an ordinary limit switch is not a safety-rated stop.

Fit a home switch and travel limits anyway #

Yes, add them — it is not optional. A home switch gives the controller a repeatable physical reference at every start-up, which no motor-shaft encoder can supply across a power cycle or a skipped tooth. A travel-limit switch marks the mechanical boundary and stops motion when the reference goes wrong; one switch can serve both roles where the controller supports it, but the functions stay distinct. Software travel limits bound motion only while the coordinate reference is correct and the mechanics intact — after a tooth jumps, they protect nothing. Mount the limit switches so they actuate before a collision can happen, and keep them in the control path so motion stops even mid-move.

The decision, case by case #

  • Shoulder and base joints at hobby motor sizes: gearbox. The worked example's 8 kg link at 300 mm needs 23.5 N·m and a 14:1 stage — past the roughly 5:1 where a single GT2 stage stays practical — so specify a planetary rated for the computed output torque at the computed ratio, from its datasheet. The all-belt route remains a real design path: a capstan-style large-ratio drive, or two rigid stages with metal pulleys and reinforced belts.
  • Elbow and wrist joints: belt, usually. Compute them the same way; most distal joints land under the belt threshold, and a belt also lets the motor sit back in the arm where its weight helps balance the link.
  • Either transmission: fit the home switch and the travel-limit switches. The magnetic encoder answers motor-shaft questions only, so the physical reference and the physical boundary are what make start-up repeatable and mistakes survivable.
  • Printed gears: prototype only. Print to prove geometry and motion, then replace any load-carrying stage with metal pulleys, reinforced belts, or a rated gearbox.

Questions asked

How far does the joint move if a GT2 belt skips one tooth?

One skipped tooth turns the joint pulley by one tooth pitch: 360 degrees divided by its tooth count. A 60-tooth joint pulley shifts 6 degrees at the joint — about 31 mm of tip motion at a 300 mm reach — while the closed-loop drive still reports the motor in position. You find out at the next home move, at a limit switch, or in a crash, which is why physical limits belong on the machine even with a closed-loop encoder.

Does a magnetic encoder keep position through a power cycle?

An AS5600-class magnetic chip reads shaft angle absolutely within one revolution, but it does not count revolutions while unpowered. After a restart the controller may know the shaft angle without knowing where the joint travelled while the machine was off, and a belt jump between sessions is invisible. Whether your drive stores multi-turn position is a drive-documentation question; design as if every power-up needs a home move.

Can I run the whole arm on printed gears instead of belts?

Builders report printed gears carrying more torque than printed belt pinions, at the cost of more backlash, and backlash grows as teeth wear through repeated reversals. It also accumulates down the kinematic chain, so the tool sees the sum of every stage. If you print a stage, support its shafts in real bearings rather than printed bores and test it in the actual load path before trusting it.

Where should the home switch sit on a joint?

Put it at a known calibration point inside the travel range — a hard stop or dowel-pin fixture, not the mechanical end of travel — and approach it from the same direction at creep speed each time, so backlash and overshoot cannot vary the reference. Set the travel-limit switches to actuate before a crash can occur. One switch can fill both roles where the controller supports the arrangement.

Keep reading

  • Selection guide

    Stepper Gearbox Encoder vs Servo Gearhead for Laser Aiming

    Convert the allowed target displacement into an angular budget before choosing hardware: displacement divided by distance, so 0.1 ft at 300 ft allows about 0.33 mrad. Backlash and loaded repeatability then decide the axis: a preloaded worm or strain-wave gearhead with an encoder on the output shaft meets it, microstepping and frame size do not.

    Motors, gearboxes and drives sizing

  • Selection guide

    Choosing Servo or Stepper for a 1% Speed Regulation Pinch Wheel

    Pick the servo: behind a 45:1 planetary reducer, a velocity-mode servo meets a 1% wheel-speed tolerance with margin, because 200-300 lb-in at 80 rpm needs only 4.9-7.4 lb-in at the motor and the loop corrects load-induced error. A stepper qualifies only when its torque-speed curve clears the load with margin at the required motor speed and a measured ripple test holds the wheel inside 0.8 rpm.

    Servo and stepper motion

  • How-to

    Match a Stepper Feed Rate to Conveyor Speed With an Encoder

    Continuously scale the conveyor encoder's measured count rate into the pulse-output velocity register: divide the count rate by the counts per inch of belt travel, then multiply by the label-feed ratio and the drive's pulses per inch of label travel. A fixed feed rate matches one belt speed only; the per-scan register update is what makes the labeler follow the belt.

    Servo and stepper motion