Backdrivable Actuator Sizing Tool & Guide

Screen a rotary joint or backdrivable linear actuator: compare reflected inertia, screw lead, drag assumptions, and rotor energy.

Actuator sizing tool

Change ratio and rotor inertia for live estimates. All defaults are illustrative; expand assumptions before comparing hardware.

Required actuator inputs; illustrative defaults

1–200 :1. Ratio of motor rotor angular speed to output joint speed. 1.0 represents direct drive.

0.0001–1000 kg·cm². Bare motor rotor inertia. (1 kg·cm² = 10⁻⁴ kg·m² = 1000 g·cm²).

Friction, electrical & stopping assumptions (11 required defaults)

0–10 N·m. No-load magnetic cogging and bearing drag measured at the bare motor shaft.

10–99 %. Mechanical efficiency when driven from output to input. Measure separately from forward efficiency; efficiency at standstill is not defined.

0–50 N·m. Static seal, grease shearing, and bearing drag measured at the joint output.

0.001–5 N·m/A. Torque per ampere of phase quadrature (q-axis) current.

0.01–100 N·m. Bare motor continuous torque under its specified speed, mounting, and cooling conditions.

0–100 rad/s. Output joint rotational velocity during external collision or ground impact.

12–400 V. Nominal supply voltage for comparison only; this is not the driver overvoltage limit.

10–100 %. Independent moving-load assumption; not inferred from reverse efficiency or valid at stall.

0.001–10 V·s/rad. Measured DC-equivalent rectified open-circuit voltage per motor mechanical rad/s. Convert phase/RMS/peak conventions first; do not copy K_t automatically.

0.1–1000 ms. Linear deceleration assumption for rotor inertial torque only; not a measured collision duration.

1–2000 mA. Current feedback noise floor including ADC noise, shunt amplifier drift, and ripple.

Illustrative screening result

Reflected rotor inertia

0.00162 kg·m²

36.0× bare rotor; excludes load and gears

Backdrive resistance estimate

0.32 N·m

Heuristic friction model; measure actual breakaway

Rotor kinetic energy

0.12 J

Stored energy, not energy dissipated or total impact energy

Average rotor stopping torque

1.94 N·m

Linear stop over 10 ms; not peak contact torque

Moving-load output torque estimate:
19.74 N·m; uses entered motor rating and forward efficiency, not a verified continuous joint rating.
Rotor speed:
688 rpm; check the actual motor speed limit.
Ideal current-noise torque contribution:
0.0264 N·m; excludes friction, observer errors, and contact geometry.
DC-equivalent open-circuit EMF:
7.92 V vs 48 V nominal. EMF is below nominal; controlled regeneration can still raise the bus voltage.
Model limits: Rotor-only dynamics; no strength, thermal, controller, or bus-energy simulation. Reverse efficiency at standstill is undefined. Do not double-count gearbox losses in both efficiency and drag. These estimates cannot establish self-locking, impact safety, sensing accuracy, or electrical protection.

Next: Measure assembled breakaway torque and verify driver energy absorption before selecting a joint.

Engineering review: . Source-specific check dates and revisions are listed below; verify the selected assembly and test conditions.

Primary sources

Key decisions before choosing a backdrivable actuator

Keep friction and inertia separate

A 6:1 ratio gives a 36× rotor inertia multiplier. Breakaway torque still needs an assembly test; a single transparency score hides this distinction.

[WENSING2017]

Do not infer self-locking from ratio

An 80:1 strain-wave component set can have finite backdriving torque. A gravity load needs a separate verified holding strategy.

[HARMONIC]

Require a regeneration energy path

Open-circuit EMF compared with nominal voltage cannot establish electrical protection. Verify drive limits, charge acceptance, and braking duty.

[BRAKING]

Measure bandwidth rather than ranking gearbox names

The 2016 ANYmal compliant joint tests report over 70 Hz torque bandwidth. Springs and high ratios alone do not set a universal limit.

[HUTTER2016]

Backdrivable linear actuators & reflected mass

In linear pushrods (ball/roller screws), reflected rotor mass scales as J_m·(2π/p)². At equal rotor inertia, changing lead from 5 mm to 2 mm multiplies reflected mass by 6.25. Check speed, load spectrum, and compliance before selecting lead.

[ROLLVIS-CATALOG]
Three independent selection checksTorque capacity, drag, and dynamic inertia must all be checked. A ratio or architecture name cannot replace the complete assembly measurements.Torque-speed dutyLoad and thermal envelopeMeasured backdriveBreakaway and running dragDynamic behaviorN²J_m + actual load model
Three independent selection checks: Torque capacity, drag, and dynamic inertia must all be checked. A ratio or architecture name cannot replace the complete assembly measurements.

1. What the tool calculates — and what needs measurement

A backdrivable actuator lets output-side torque move the motor through the transmission in a specified electrical state. This page covers the complete motor, transmission, driver, and sensing package. For measured torque screening, reducer terminology, and backdrivable gearbox selection rules, see the backdrivable gearbox & backdrivability guide.

J_ref = N²J_m; ω_motor = Nω_joint
τ_resistance,estimate = Nτ_drag,motor / η_rev + τ_drag,gearbox
τ_output,estimate = Nτ_motor,ratedη_fwd

Rotor inertia is referred through an ideal rigid ratio. The resistance expression is a heuristic: efficiency is undefined at standstill, and gearbox drag must exclude losses already included in η_rev. Forward efficiency is entered independently. The output-torque expression applies to an assumed moving operating point; it does not model a joint thermal rating or stall losses. All fields are required, with example defaults and stated input ranges.

Reverse mechanical power pathOutput motion drives the reducer and motor. Resistance depends on load, lubrication, and electrical state; finite motion is not a safety result.Output jointExternal torque and speedTransmissionFriction + elastic deflectionMotor / driverRotor inertia + current state
Reverse mechanical power path: Output motion drives the reducer and motor. Resistance depends on load, lubrication, and electrical state; finite motion is not a safety result.

Record measured breakaway torque and running drag at multiple speeds, temperatures, loads, and drive states. Do not equate the input starting torque in a reducer catalog with output backdriving torque. [HARMONIC-CATALOG]

2. Reflected inertia, stored energy & stopping assumptions

E_rotor = ½N²J_mω_joint²
τ_rotor,average = J_ref · Δω_joint / Δt
1 kg·cm² = 10⁻⁴ kg·m²; 1 g·cm² = 10⁻⁷ kg·m²

The model describes stored rotor energy and average rotor inertial torque under a linear stop. It omits links, payload, gearing, elasticity, and external work. It cannot estimate peak tooth stress, contact force, or energy dissipated in an impact. [WENSING2017]

Ratio squared: rotor inertia multiplierExact kinematic multipliers for the same rotor: 1:1 → 1×, 6:1 → 36×, 16:1 → 256×, 100:1 → 10,000×. Bars use a logarithmic scale; no failure threshold is implied.1:11× J_m6:136× J_m16:1256× J_m100:110,000× J_m
Ratio squared: rotor inertia multiplier: Exact kinematic multipliers for the same rotor: 1:1 → 1×, 6:1 → 36×, 16:1 → 256×, 100:1 → 10,000×. Bars use a logarithmic scale; no failure threshold is implied.
Illustrative same-rotor comparison: J_m = 0.45 kg·cm²; ω_joint = 12 rad/s; no manufacturer data
Ratio NN² multiplierJ_ref (kg·m²)Rotor energy (J)
1:11×0.0000450.00324
6:136×0.001620.11664
16:1256×0.011520.82944
100:110,000×0.4532.4
Assumed stopping time changes average rotor torqueIllustrative rotor-only model: J_ref = 0.00162 kg·m² and initial speed 12 rad/s. Average inertial torque is 3.888 N·m at 5 ms, 1.944 N·m at 10 ms, and 0.972 N·m at 20 ms; actual contact peaks are unknown.5 ms3.888 N·m10 ms1.944 N·m20 ms0.972 N·m
Assumed stopping time changes average rotor torque: Illustrative rotor-only model: J_ref = 0.00162 kg·m² and initial speed 12 rad/s. Average inertial torque is 3.888 N·m at 5 ms, 1.944 N·m at 10 ms, and 0.972 N·m at 20 ms; actual contact peaks are unknown.

3. Architecture tradeoffs & application fit

Use these qualitative engineering comparisons to shortlist candidates. Ratios below are examples, not class boundaries. No architecture alone guarantees impact tolerance, human contact safety, low drag, or a particular force bandwidth. [KATZ2018][HUTTER2016]

Five actuator architectures; complete-assembly measurements govern selection
ArchitectureExample ratioPotential benefitRequired verificationCandidate use
Direct drive1:1No gearbox losses; motor drag remainsRotor inertia and continuous holding heatHaptics or joints with a suitable motor envelope
QDD / low-ratio planetary6:1 exampleLow ratio can reduce referred inertiaWinding heat, real drag, and driver energy pathDynamic contact, subject to torque-speed duty
Multi-stage planetary16:1 exampleMore torque leverage in a compact motorStage losses, backlash, and bearing loadsPayload-driven joints with measured drag limits
Strain wave80:1 exampleCompact reduction and low backlashBackdrive torque, nonlinear stiffness, and fatiguePrecision positioning with system-level force control
Series elastic actuatorRatio depends on designDeflection sensing and mechanical complianceSpring travel, resonance, control, and packagingContact tasks needing physical elasticity
Decision sequence before procurementSet limits from the application, compare complete assemblies, and then accept hardware against the agreed test plan. Unknown supplier data remains a procurement question.Define task limitsDrag, duty, speed, loadsCompare candidatesComplete assembly evidenceVerify on benchMeasured acceptance gates
Decision sequence before procurement: Set limits from the application, compare complete assemblies, and then accept hardware against the agreed test plan. Unknown supplier data remains a procurement question.

3b. Backdrivable linear actuators: roller screws, ball screws & reflected mass

A backdrivable linear actuator allows an external axial load to move the screw and motor in a specified brake and drive state. Screw lead p is axial travel per screw revolution, not thread pitch on a multi-start screw. This calculator assumes a rigid, directly coupled motor and screw: ω_motor = (2π/p)v. A belt or gearbox adds its ratio; a series spring requires a dynamic model. The cited catalog supports lead and efficiency terminology; reflected mass below is derived kinematically for this tool. Use the linear calculator in the tool selector above.[ROLLVIS-CATALOG]

Linear reverse force transmission and self-locking thresholdFor an ideal square-thread sliding model, λ > φ = arctan(μ) permits backdriving; λ ≤ φ indicates self-locking. Trapezoidal thread angle, rolling contact, preload, brakes, and drive state require separate verification.Axial Contact ForceF_ext on pushrod / nutScrew Lead Angleλ > φ allows backdriveReverse Motor Torqueτ_bd = p·F_ext·η_rev / 2π
Linear reverse force transmission and self-locking threshold: For an ideal square-thread sliding model, λ > φ = arctan(μ) permits backdriving; λ ≤ φ indicates self-locking. Trapezoidal thread angle, rolling contact, preload, brakes, and drive state require separate verification.

Governing mechanics of linear backdrivability

Self-locking boundary

In an ideal square-thread sliding-friction model, λ = arctan(p/(πd_mean)) and φ = arctan(μ). The moving efficiencies are η_fwd = tan(λ)/tan(λ + φ) and η_rev = tan(λ − φ)/tan(λ). λ ≤ φ indicates self-locking in that model. Trapezoidal threads need a thread-angle friction correction; rolling screws require manufacturer geometry and loss data. Sliding lead screws can backdrive at sufficiently high leads; their name alone does not establish self-locking. Confirm the actual assembly, including preload, alignment, and brake state.[THOMSON-LEAD-SCREWS][EWELLIX-ROLLER-SCREWS]

Linear reflected rotor mass

m_ref = J_rotor · (2π / p)²

With illustrative J_m = 0.45 kg·cm², 5 mm lead gives 71.06 kg equivalent rotor mass; 2 mm gives 444.13 kg, a 6.25× increase. These are kinematic equivalents, not physical payload mass or measured impact loads. Include screw and coupler inertia, rod, payload, and linkage geometry in a complete model.

Approximate backdrive resistance

F_resistance,estimate = 2π · τ_motor,drag / (p · η_rev) + F_extra,drag

This is a moving-loss heuristic, not a static breakaway prediction: efficiency is undefined at standstill. Enter extra nut, seal, and guide drag only if it is excluded from the efficiency assumption, to avoid double-counting. Forward thrust subtracts entered motor drag from the electromagnetic torque assumption, applies forward efficiency, then subtracts extra output drag, with zero available thrust if losses consume it. If entering net shaft torque, add the motor drag already excluded to the torque input and retain the measured motor drag for reverse screening. No universal seal or preload force is assumed. Measure assembled start-up and running force in both directions with a calibrated in-line load cell.[EWELLIX-ROLLER-SCREWS]

Average stopping force and load-rating limits

F_rotor,average = m_ref · Δv / Δt

For the illustrative 2 mm screw at 0.5 m/s stopping in 10 ms, the rotor-only average is 22.21 kN. It does not predict peak foot-ground force, raceway stress, or damage. Dynamic load rating C_a is a fatigue-life reference; static rating C_0a is a separate permanent-deformation reference. Neither is an allowable impact force without the supplier load and support model. There is no universal minimum lead for a leg actuator: check stroke, buckling, critical speed, motor RPM, duty, and measured impact response.[ROLLVIS-CATALOG]

Linear actuator calculator with all thirteen example inputs and results: 71.06 kg reflected rotor mass, 52.0 N moving resistance and 3,553.1 N average stopping force
Actual software capture, 2026-11-08; no physical actuator test, customer result or supplier endorsement. Shared sticky navigation is hidden in this tool-only capture. Select Linear screw and Reset linear defaults. Inputs: lead 5 mm/rev; rotor inertia 0.45 kg·cm²; motor drag 0.025 N·m; reverse efficiency 85%; extra output drag 15 N; K_t 0.11 N·m/A; assumed motor torque 3.5 N·m; speed 0.5 m/s; nominal bus 48 V; current noise 20 mA; forward efficiency 92%; K_e 0.11 V/(rad/s); stop duration 10 ms. Reproduced outputs: 71.06 kg reflected rotor mass, 52.0 N moving resistance, 4,002.5 N moving-load thrust, 8.88 J rotor energy, 3,553.1 N average stopping force, 6,000 RPM, 69.12 V open-circuit EMF and 2.76 N ideal current-noise contribution. These are illustrative calculations with the model limits stated above.
Reflected linear mass scales with (2π/p)²Illustrative directly coupled motor rotor J_m = 0.45 kg·cm² (4.5×10⁻⁵ kg·m²). Linear bar widths show equivalent rotor mass: 10 mm lead = 17.8 kg, 5 mm = 71.1 kg, 2 mm = 444.1 kg. These kinematic equivalents do not predict peak impact force.10 mm lead17.8 kg5 mm lead71.1 kg2 mm lead444.1 kg
Reflected linear mass scales with (2π/p)²: Illustrative directly coupled motor rotor J_m = 0.45 kg·cm² (4.5×10⁻⁵ kg·m²). Linear bar widths show equivalent rotor mass: 10 mm lead = 17.8 kg, 5 mm = 71.1 kg, 2 mm = 444.1 kg. These kinematic equivalents do not predict peak impact force.
Qualitative linear architecture screening; efficiencies and load limits require the selected assembly's data
Transmission typeLead selectionForward lossesReverse behaviorReflected mass modelLoad verificationForce sensing limitsCandidate use
Precision Ground Ball ScrewSelect lead for speed, thrust, and inertiaRolling losses; preload and lubrication matterOften backdrivable; test assembled dragJ_m·(2π/p)², plus screw and moving massCheck nut ratings, bearings, buckling, and speedCurrent observer needs friction and dynamics calibrationPushrods with suitable load and stroke duty
Planetary Roller Screw (PRS/IPRS)Select actual series and nut configurationUse model-specific efficiency; no family-wide percentageCheck reversibility and holding-brake requirementsSame rigid screw model; lead controls multiplierLoad capacity is model-specific, not an impact ratingPreload, seals, and guides affect observer errorHigh axial-load packaging subject to measured duty
Acme / Trapezoidal Sliding Lead ScrewLead and friction determine reversibilitySliding friction; verify thermal dutyCan self-lock at low leads; high leads may backdriveRigid kinematic model only while movingCheck wear, material, buckling, and load spectrumHysteresis can limit current-based estimatesPositioning or holding tasks after verification
Direct Drive Linear Motor (Tubular/Slotless)No screw leadNo screw losses; winding and guide losses remainGuide, cogging, and electrical state govern dragNo screw-reflected rotor term; moving mass remainsGuides, mounts, and end stops still carry impact loadsForce constant, ripple, dynamics, and sensor calibration matterDirect force control within thermal and package limits
Series Elastic Linear Actuator (SELA)Transmission plus selected springDepends on transmission and spring lossesCompliance is not an efficiency valueFrequency-dependent coupling; rotor inertia remainsCheck spring travel, bottoming, fatigue, and resonancesDeflection sensing needs stiffness and hysteresis calibrationContact tasks requiring designed physical compliance

Screw comparisons: [ROLLVIS-CATALOG][THOMSON-LEAD-SCREWS]. Direct-drive and elastic rows are engineering screening considerations, not measured product rankings. A spring changes dynamic coupling; it does not remove motor inertia or guarantee impact survival.

4. Motor geometry: torque, rotor inertia & heat

For a radial-flux motor with uniform tangential air-gap shear stress σ and active length L, T = 2πσr²L. This is a scaling relation under fixed loading assumptions, not proof of a fixed torque-density advantage. A larger radius can increase rotor inertia as well as torque. Compare the complete motor-reducer pair at the required duty. [SEOK2015]

Motor geometry comparison conditionsAt equal air-gap shear stress and active axial length, radial-flux torque scales with radius squared. Package torque density and rotor inertia require actual geometry and thermal data.Air-gap shearAssume equal loadingRadius and lengthT = 2πσr²LComplete jointCheck mass, heat, N²J_m
Motor geometry comparison conditions: At equal air-gap shear stress and active axial length, radial-flux torque scales with radius squared. Package torque density and rotor inertia require actual geometry and thermal data.
Topology tradeoffs are qualitative; no universal torque-density multiplier is asserted
Motor geometryOpportunityTradeoffSelection evidence
Large-radius radial-flux motorTorque scales with air-gap surface shear and moment armA larger radius can also raise rotor inertiaCheck measured K_m, winding temperature, and package inertia
Compact inrunnerCan pair low rotor inertia with higher speedReduction needed for output torque raises N²J_mCompare complete motor + reducer inertia and loss
Axial-flux motorShort axial package may fit flat jointsThermal path and structural deflection depend on designRequire thermal and torque-speed data, not topology claims

5. Manufacturer evidence: known values & procurement gaps

Rotary sources checked 2026-10-27; linear references rechecked 2026-11-08. Linear rows identify product families, not selected models. These are manufacturer claims for specific revisions, not independent tests or an apples-to-apples ranking. A reducer component set or bare roller screw excludes the motor and driver present in an integrated module. Peak torque is not continuous duty. Screw load-capacity references are not motor thrust ratings or impact test results.

Revision-specific published values; N/A means the cited page does not establish the required input
Product & sourceRatio / screw leadPublished torque / load referenceBackdrive evidenceMassUnknown or to confirm
CubeMars AK80-9 V3.0 KV100
[CUBEMARS]
9:19 N·m rated / 22 N·m peak0.51 N·m (catalog back drive)490 gBreakaway test conditions and thermal mounting; listed rotor inertia 1118.3238 g·cm² needs revision/reference confirmation. No derived inertia ranking.
Unitree GO-M8010-6
[UNITREE]
6.33:123.7 N·m maximum; continuous rating N/A in cited pageN/A — not published on cited pageAbout 530 gRotor inertia, breakaway torque, and continuous thermal duty; do not substitute estimated values.
Harmonic Drive CSG-20-80-2A-GR component set
[HARMONIC]
80:144 N·m rated L10 / 96 N·m repeated peak4.6 N·m (output backdriving); input starting torque is a separate 4.5 N·cm0.28 kg (component set only)Motor, driver, mounting bearings, and assembled system thermal/drag behavior are not included.
Ewellix roller-screw family (not a selected model)
[EWELLIX-ROLLER-SCREWS]
Model-specific screw leadC_a / C_0a are life / static-load references, not motor thrust ratingsN/A — assembled drag not establishedN/A — shaft length and configuration requiredSelect a nut/shaft configuration and verify preload, screw speed, supports, thermal duty, and motor integration; no SR 12×5 assembly claim is inferred.
Rollvis RV / HRV / RVI roller-screw families
[ROLLVIS-CATALOG]
Lead differs from thread pitch for multi-start screwsModel-specific tables; fatigue capacity is not impact toleranceN/A — load-cell breakaway test requiredN/A — integrated assembly unspecifiedConfirm exact series, lead, nut type, preload, and supports. The cited catalogue does not establish complete assembled actuator mass or measured drag.
Schaeffler PWG family (2017 brochure)
[SCHAEFFLER-PWG]
Family small-lead range 0.7–3 mmFamily dynamic capacity 8–40 kN across spindle diameters 5–25 mmN/A — no breakaway force established by brochureN/A — selected assembly not specifiedFamily ranges cannot be assigned to a PWG 10×1 model. Confirm exact lead, load ratings, reversibility, motor, and bearing integration.

The cited AK80-9 V3.0 page currently lists rotor inertia as 1118.3238 g·cm². Earlier material and other revisions use different values; request the supplied revision and reference plane before deriving reflected inertia. The GO-M8010-6 page does not provide a comparable breakaway or rotor-inertia test. Leave those inputs unknown until the supplier confirms them. Similarly, roller screw catalog entries (Ewellix SR, Rollvis RV, Schaeffler PWG) specify spindle and nut mechanical ratings, but exclude the customer-integrated frameless motor rotor, bearing preload, and pushrod wiper seals; always measure the assembled linear actuator on an instrumented load-cell test bench.

6. Current feedback estimates motor torque, not contact force

τ_motor ≈ K_t I_q
δτ_current,ideal = N K_t δI_q

The displayed noise contribution omits friction, current-loop bias, cogging, gravity, inertial torque, and filtering. A joint torque in N·m becomes contact force in N only through known contact geometry and a suitable Jacobian. Calibrate an external torque observer across angle, speed, load, and temperature before using its output for control. [WENSING2017][MAXON]

From current to an external torque estimateMotor torque is only the first step. Subtract inertial and friction effects and validate the output observer against a calibrated torque measurement.Calibrated currentτ_motor ≈ K_t I_qDynamics observerInertia, gravity, frictionReference testError vs load and speed
From current to an external torque estimate: Motor torque is only the first step. Subtract inertial and friction effects and validate the output observer against a calibrated torque measurement.
Bandwidth is an assembly measurementNo universal Hz ranking follows from reducer type. Identify resonances, measure response under the intended load, and report the measurement method and amplitude.Identify mechanicsFriction and stiffnessClose the control loopSensor + delay + loadMeasure responseBandwidth + stability margin
Bandwidth is an assembly measurement: No universal Hz ranking follows from reducer type. Identify resonances, measure response under the intended load, and report the measurement method and amplitude.

For observer design and control tradeoffs, continue with force control in robotics and impedance control.

7. Dual encoders: torsional deflection & uncertainty

Δθ = θ_output − θ_motor/N
τ_elastic,estimate = K_torsional Δθ (sign defined by calibration)

Synchronize and unwrap both encoder angles, remove zero offsets, and identify stiffness with known bidirectional loads. Encoder quantization, offsets, backlash, nonlinear stiffness, and hysteresis can dominate a small deflection signal. Use an error budget and reference torque calibration; two encoders do not automatically make a torque sensor. The Harmonic Drive product data, for example, lists multiple stiffness regimes. [HARMONIC]

8. Regeneration: separate EMF comparison from energy absorption

Use a separately measured K_e with a documented voltage and speed convention. Here it means DC-equivalent rectified open-circuit voltage per motor mechanical rad/s. K_t and K_e are numerically interchangeable only with compatible definitions. EMF = K_eN|ω_joint| is a nominal comparison, not a bus-voltage prediction.

An active inverter can regenerate below the nominal bus voltage. In an unpowered but connected drive, rectifier paths depend on the bridge and electrical state. Establish the actual energy path, voltage/current ratings, and supply charge acceptance; a TVS for short transients is not a substitute for a rated repetitive braking sink.

Regeneration needs an energy destinationA backdriven motor and active inverter can deliver energy to the bus. Verify supply charge acceptance or a rated dump path; capacitance stores energy temporarily.Mechanical energyRotor + load + gravityMotor and inverterElectrical state mattersDC bus energy sinkSupply / battery / chopper
Regeneration needs an energy destination: A backdriven motor and active inverter can deliver energy to the bus. Verify supply charge acceptance or a rated dump path; capacitance stores energy temporarily.
Ideal electrical balances and bounds; use the actual driver manual, tolerances, and duty ratings
QuantityFormula / boundBoundary or required action
Rotor stored energyE_rotor = ½N²J_mω²One contribution to energy; add links, payload, gravity, and continuing external work.
Capacitor-only energy balanceC ≥ 2E_bus / (V_max² − V_initial²)Requires V_max > V_initial and a defined net E_bus. Include simultaneous axes, voltage tolerance, ESR, and ripple; not a complete protection design.
Chopper current lower boundR ≥ V_bus,max / I_switch,maxAlso obey the drive minimum resistance and component tolerances. An undersized resistance can overcurrent the switch.
Braking power upper boundR ≤ V_on² / P_requiredCheck available duty, peak/pulse energy, average heat, wiring, and thermal cutout. If bounds conflict, change the drive or energy sink.

The current limit sets a lower resistance bound; power absorption sets an upper bound. Evaluate current at the highest allowed bus voltage, and include pulse and average thermal limits. The general selection logic is documented by Schneider Electric; its industrial drive voltage examples are not robot-driver recommendations. [BRAKING]

9. Thermal duty & the motor constant

K_m = K_t / √R; P_copper = τ_motor² / K_m²

This compact expression requires matched torque, current, and resistance conventions, including phase factors for three-phase motors. K_m has units N·m/√W. It is not torque per watt. Account for speed-dependent iron loss, inverter loss, ambient temperature, mounting conduction, and torque-speed-time duty. The calculator uses your entered motor rating; it does not simulate cooling or prescribe a universal derating percentage. [SEOK2015][MAXON]

10. Turn the result into an actuator acceptance plan

Before requesting a quote, specify torque-speed duty, load inertia, drag limits, thermal mounting, electrical states, bearing loads, envelope, and regeneration behavior. Set acceptance limits from your application and the supplier manual. The following plan is an engineering proposal, not a published industry threshold.

Suggested pre-purchase measurements; all pass criteria must be agreed for the actual assembly
GateTest evidenceAcceptance basisRisk if omitted
Backdrive resistanceMeasure both directions, cold/warm, with documented drive/brake statesApplication-defined maximum breakaway and running drag; include uncertaintyUnexpected resistance or unstable torque estimation
RegenerationInstrument bus voltage/current during representative deceleration and backdrivingSupplier voltage/current limits, clamp thresholds, pulse energy, and supply charge acceptanceTrips, component overload, or loss of torque control
Thermal dutyReproduce torque-speed-time cycle and actual mounting/cooling to equilibriumVendor winding, magnet, bearing, and driver limits with agreed marginHolding-torque derating or reduced component life
Sensing and stiffnessApply known bidirectional loads; compare current and encoder estimates with reference torqueApplication error budget over angle, load, temperature, and bandwidthObserver bias, hysteresis, or position/torque loop instability
Mechanical load survivalUse a guarded, instrumented fixture and an agreed joint load spectrumSupplier bearing, gear, fatigue, and alignment limits; inspect after testUnrepresentative drop tests or undetected structural damage

Discuss joint requirements

11. Bench states & a reproducible backdrive protocol

Guarded backdrive test fixtureRestrain gravity loads, measure torque at the output, and record motion and temperature. Open phases, connected driver, and active control are distinct test states.Guarded fixtureRestraint + calibrated loadOutput transducerTorque and encoder motionSynchronized loggingTemperature + drive state
Guarded backdrive test fixture: Restrain gravity loads, measure torque at the output, and record motion and temperature. Open phases, connected driver, and active control are distinct test states.
  1. Restrain gravity loads and use a guarded fixture with a calibrated output torque transducer and encoder.
  2. Document the exact electrical state: mechanically disconnected motor, open phases, connected unpowered driver, or active torque control. Follow the driver manual; do not improvise phase shorting.
  3. Ramp output torque slowly in both directions. Record onset of motion, running drag versus speed, lubricant and winding temperature, preload, and mounting conditions.
  4. Repeat under the intended load and thermal envelope. Report uncertainty and compare against the agreed application limit.

The catalog backdriving values use stated reference conditions and cannot be assumed to hold for every assembled joint. [HARMONIC-CATALOG]

12. Worked scenarios & published system evidence

The four calculations below are illustrative scenarios, not customer results or measured product performance. Their assumptions and arithmetic can be reproduced independently.

Illustrative ratio tradeoff

Assumptions: Same rotor J_m = 0.45 kg·cm² = 4.5×10⁻⁵ kg·m²; joint speed 12 rad/s. Ignore transmission and load inertia.

Method: Compare N = 6 and N = 16 using J_ref = N²J_m and E = ½J_refω².

Result & limit: 6:1 gives 0.00162 kg·m² and 0.11664 J; 16:1 gives 0.01152 kg·m² and 0.82944 J. At equal motor torque the higher ratio offers more leverage, but these values do not predict impact strength.

Illustrative stopping-time sensitivity

Assumptions: Use the 6:1 example above at 12 rad/s; assume linear deceleration with no external load contribution.

Method: Calculate average rotor inertial torque J_refω/Δt for 10 ms and 5 ms.

Result & limit: The estimates are 1.944 N·m and 3.888 N·m. The doubled average is a model sensitivity result, not a peak force or tooth-stress test.

Illustrative capacitor-only energy budget

Assumptions: Assume net energy delivered to the bus is 1 J, initial bus 48 V, chosen maximum 60 V, no other storage or sink.

Method: Use C ≥ 2×1 / (60²−48²); the energy assumption is independent of mechanical reverse efficiency.

Result & limit: The ideal minimum is 0.001543 F, approximately 1543 µF. This does not select a real capacitor bank: tolerances, ESR, repetition, driver limits, and supply state require verification.

Illustrative screw lead: reflected mass & rotor energy

Assumptions: A humanoid robot knee pushrod driven by a frameless motor (J_m = 0.45 kg·cm² = 4.5×10⁻⁵ kg·m²) operates at 0.5 m/s impact speed. Compare 5 mm and 2 mm leads at the same rotor inertia, independently of screw family.

Method: Use m_ref = J_m · (2π/p)². For 5 mm lead: 2π/0.005 ≈ 1256.6 rad/m → m_ref ≈ 71.06 kg. For 2 mm lead: 2π/0.002 ≈ 3141.6 rad/m → m_ref ≈ 444.13 kg. Kinetic energy E = ½m_ref v².

Result & limit: Dropping screw lead from 5 mm to 2 mm increases reflected rotor mass by (5/2)² = 6.25× (from 71.06 kg to 444.13 kg), increasing stored rotor kinetic energy from 8.88 J to 55.52 J. This comparison isolates lead at equal rotor inertia; it does not rank screw families or predict structural damage. Check critical screw speed, load spectrum, supports, and available compliance before selecting a lead.

Backdrivable actuator tool showing 0.00162 kg·m² reflected rotor inertia and 1.94 N·m average stopping torque for the illustrative 6:1 example
Actual software capture, 2026-10-27; no physical actuator test or customer result. Reproduce with Restore example defaults: N = 6, rotor inertia = 0.45 kg·cm², motor drag = 0.025 N·m, gearbox drag = 0.15 N·m, reverse/forward efficiency = 88%/94%, K_t = 0.11 N·m/A, motor rating = 3.5 N·m, joint speed = 12 rad/s, nominal bus = 48 V, K_e = 0.11 V·s/rad, stopping time = 10 ms, current noise = 40 mA. Exact rotor energy is 0.11664 J; the screen rounds it to 0.12 J.

Published low-ratio design evidence: Katz’s 2018 MIT thesis documents modular actuator hardware and controls for a small quadruped, with additional teleoperation and biped applications. It supports using an integrated motor, transmission, and control design process; it does not validate the tool presets as commercial product specifications. [KATZ2018]

Evidence against a universal SEA bandwidth ceiling: Hutter et al. report torque-control bandwidth above 70 Hz for the 2016 ANYmal compliant joint modules. That result belongs to the tested assembly and controller, not every series elastic actuator. [HUTTER2016]

13. Misuse, cost & scenario mismatch

Engineering risk assessment prompts and practical alternative paths
RiskWhy it mattersMitigation or alternative
Misuse: inferred safetyA small rotor energy estimate excludes link and payload inertiaMeasure system contact/stopping behavior; add passive compliance or lower speeds where needed
Cost: incomplete module comparisonReducer-only prices omit motor, driver, bearings, calibration, and coolingCompare installed system cost and service effort on the same duty cycle
Scenario mismatch: long holding dutyA low ratio can increase required motor current and heatVerify equilibrium duty; evaluate a larger motor, alternative ratio, or a separate holding brake
Power loss: assumed holdingA backdrivable joint can release a gravity load when torque is removedProvide a load restraint or rated holding brake; validate the power-loss state

These are application review prompts. Backdrivability is not a substitute for a system risk assessment, load restraint, or validated contact behavior. [ISOTS15066]

14. Frequently asked actuator selection questions

Physics & Model Limits

Evidence and further reading: [WENSING2017][HARMONIC-CATALOG][ROLLVIS-CATALOG]

What defines a backdrivable actuator or backdrivable linear actuator?

External force or torque applied at the output link drives the transmission (rotary gearbox or linear screw) and motor backward in a specified electrical state. Breakaway force/torque, running drag, speed, load, and temperature determine whether that backdriving motion is useful for dynamic compliance or haptic interaction.

Why does rotor inertia scale with N² for rotary joints and (2π/p)² for linear actuators?

For rotary actuators, the rotor spins at N times joint speed, giving an output-referred inertia N²J_motor. For a linear actuator with screw lead p, angular speed is ω = (2π/p)·v, yielding an equivalent reflected linear mass m_ref = J_motor·(2π/p)². Fine screw leads drastically multiply reflected mass.

Is a transparency percentage a standard actuator rating?

No universal percentage is used here. Dividing breakaway torque by rated torque can conceal a large absolute resistance and ignores dynamic inertia. Compare measured drag and reflected inertia separately against application limits.

Can reverse efficiency predict static breakaway torque?

Only as a rough loss assumption. Efficiency is a power ratio and is undefined at zero speed. Preload, lubricant, seals, temperature, and electrical state require a separate assembled breakaway test.

Does high reduction mean that a joint is self-locking?

No. High ratios increase the rotor inertia multiplier but do not prove self-locking. Harmonic Drive publishes finite backdriving torque values and advises using a fail-safe brake for holding loads.

Choosing a Joint Architecture

Evidence and further reading: [KATZ2018][HUTTER2016][THOMSON-LEAD-SCREWS]

When should I shortlist QDD?

Shortlist it when low output drag and low reflected inertia support dynamic contact tasks. Check motor thermal duty, peak torque duration, gear strength, bearing loads, and driver regeneration before choosing a ratio.

Does series elasticity always limit control bandwidth to 20 Hz?

No. Bandwidth depends on actuator mechanics, load, sensing, controller, and measurement method. The 2016 ANYmal paper reports torque control bandwidth above 70 Hz for its compliant joint modules.

Why consider direct drive?

It removes gearbox friction and backlash. Rotor inertia, bearing friction, cogging, mounting stiffness, and winding heat remain. Evaluate motor size and thermal duty at the required output torque.

When is a backdrivable linear actuator preferred over a rotary actuator?

Consider a linear actuator when the pushrod stroke and packaging fit the joint kinematics. Compare axial duty, linkage leverage throughout travel, screw speed, buckling, friction, and reflected mass. A catalog load rating alone does not establish survival of ground impacts.

Are planetary roller screws and ball screws backdrivable compared to lead screws?

Many ball and roller screws can be backdriven, but lead, preload, lubrication, alignment, and electrical/brake state matter. Sliding acme screws can self-lock at low leads and can backdrive at sufficiently high leads. Use manufacturer data and an assembled bidirectional test; do not assign a universal reverse efficiency by screw family.

Sensing & Electrical Behavior

Evidence and further reading: [WENSING2017][MAXON][BRAKING]

Can motor current estimate external joint torque or linear actuator force?

Current estimates electromagnetic torque through K_t. For linear actuators, thrust is F = (2π·K_t·I_q·η_fwd)/p. Estimating external contact wrench requires subtracting rotor inertia, screw mass, seal friction, and nut drag; friction and observer calibration limit the accuracy of current-based external-force estimates.

Is N·K_t·current noise the minimum detectable contact torque?

No. It is only an ideal motor-current noise contribution referred to the output. Actual detection also depends on friction, torque ripple, calibration, filtering, observer bias, and contact geometry.

Can K_t be copied directly into the K_e field?

Only after confirming compatible definitions. Phase versus line-to-line, RMS versus peak, and mechanical versus electrical speed conventions can introduce conversion factors. The tool requires a separate DC-equivalent measured K_e.

Is a back-EMF estimate below nominal bus voltage reassuring?

It does not establish protection. An active inverter can regenerate energy even when open-circuit EMF is below bus voltage. Supply charge acceptance, bus capacitance, clamp behavior, and actual overvoltage limits still need verification.

Can dual encoders replace a torque sensor?

They can support a calibrated elastic-deflection observer. Angle synchronization, zero offsets, stiffness variation, backlash, and hysteresis limit accuracy. Validate against a reference transducer before relying on the observer.

Verification & Procurement

Evidence and further reading: [HARMONIC-CATALOG][BRAKING][ISOTS15066]

How should breakaway torque or linear breakaway force be measured?

Restrain the load and use a calibrated output torque transducer or in-line load cell. Record the electrical state, temperature, direction, speed ramp, and mounting. Measure both start-up resistance and running drag across the required stroke and velocity range.

What does the assumed stopping time change?

The rotor-only average inertial torque is J_ref·Δω/Δt (or linear force m_ref·Δv/Δt). Halving the assumed time doubles that value. Actual peak joint and contact loads require a load model or instrumented test; the tool does not predict them.

How should a brake resistor be selected?

Respect the actual drive minimum resistance and peak current limits, then ensure resistance is low enough to absorb the required power. Check pulse energy, repetition rate, resistor tolerance, thermal protection, and fault behavior.

Does low backdrive torque establish collaborative robot safety?

No. Contact geometry, speed, payload, trapping, stopping behavior, and system risk assessment remain relevant. This calculator does not test conformity with ISO/TS 15066 or any other safety standard.

What should I send with an actuator RFQ?

Send torque-speed and duty cycles, load inertia, mounting and cooling conditions, acceptable measured drag, supply and regeneration limits, bearing loads, envelope, electrical states, and acceptance test requirements.

15. Primary sources, scope & review date

Responsible publisher: Humanoid Joint. First recorded: (source history; public publication date unverified). Reviewed . Next evidence review due . This is an engineering screening guide with source-backed facts, explicitly illustrative calculations, and proposed procurement checks. Unknown product values remain unknown. It does not certify hardware, predict injury, or validate a control system. Recheck manufacturer revisions and current standards before design approval.

[WENSING2017] Proprioceptive Actuator Design in the MIT Cheetah: Impact Mitigation and High-Bandwidth Physical Interaction for Dynamic Legged Robots

P. M. Wensing, A. Wang, S. Seok, D. Otten, J. Lang, and S. Kim · IEEE Transactions on Robotics; DOI 10.1109/TRO.2016.2640183 · 2017 · Checked 2026-10-27

Research basis for low reflected inertia and current-based torque estimation; no universal collision or sensing threshold is implied.

[SEOK2015] Design Principles for Energy-Efficient Legged Locomotion and Implementation on the MIT Cheetah Robot

S. Seok et al. · IEEE/ASME Transactions on Mechatronics 20(3), 1117–1129 · 2015 · Checked 2026-10-27

Motor geometry, transmission loss, regenerative electronics, and thermal loss must be considered together.

[KATZ2018] A Low Cost Modular Actuator for Dynamic Robots

Benjamin G. Katz; advisor Sangbae Kim · MIT S.M. thesis, Mechanical Engineering · 2018 · Checked 2026-10-27

Open research reference for actuator hardware and control integration; not a datasheet for commercial AK-series motors.

[HUTTER2016] ANYmal – A Highly Mobile and Dynamic Quadrupedal Robot

M. Hutter et al. · IROS 2016, 38–44; ETH accepted manuscript · 2016 · Checked 2026-10-27

Compliant joint modules achieved over 70 Hz torque control bandwidth in the reported tests; series elasticity does not imply a universal low bandwidth ceiling.

[CUBEMARS] AK80-9 V3.0 KV100 – Official Specifications

CubeMars · Manufacturer product page, Specification table · undated · Checked 2026-10-27

Revision-specific 9:1 module specifications. Confirm thermal rating conditions and the inertia reference before using catalog data in a design.

[UNITREE] GO-M8010-6 – Bionic Robot Joint Motor

Unitree Robotics · Manufacturer product parameters · undated · Checked 2026-10-27

Publishes ratio, maximum torque, speed, and mass. Does not establish breakaway torque or rotor inertia for this screening model.

[HARMONIC] CSG-20-80-2A-GR – Gear Performance Data

Harmonic Drive LLC · Manufacturer component-set specifications · undated · Checked 2026-10-27

80:1 component set with finite specified backdriving torque; not a complete motor/driver module or a holding brake.

[HARMONIC-CATALOG] CSF / CSG Catalog – Starting Torque and Backdriving Torque

Harmonic Drive LLC · Engineering catalog, printed p. 48 (PDF p. 48), tables 36–39 · undated · Checked 2026-10-27

Separates input starting torque from output backdriving torque; specifies 20°C reference conditions and advises a fail-safe brake for holding.

[BRAKING] Criteria for Braking Resistor Selection

Schneider Electric · Technical FAQ FA376712; updated 2026-04-22 · 2026 · Checked 2026-10-27

Minimum resistance protects the switching transistor; maximum resistance limits available braking power. Actual drive limits and duty ratings govern selection.

[MAXON] Formulae Handbook – Motor Sizing and Operating Cycles

maxon · Motor engineering handbook · undated · Checked 2026-10-27

Motor current, torque constants, and operating-cycle conventions must be consistent when estimating heating and torque.

[ISOTS15066] ISO/TS 15066:2016 – Robots and Robotic Devices: Collaborative Robots

International Organization for Standardization · Technical specification; status checked 2026-10-27 · 2016 · Checked 2026-10-27

Applies to collaborative industrial robot systems, not non-industrial robots; its principles may inform other robotics. Mechanical backdrivability alone does not establish safety or conformity.

[EWELLIX-ROLLER-SCREWS] Roller Screws — Engineering Catalogue IL-05003/1-EN

Ewellix · June 2020 catalogue, pp. 22–25 and calculation section; distributor-hosted manufacturer PDF · 2020 · Checked 2026-11-08

Distinguishes practical efficiency, breakaway torque, back-driving, and static versus dynamic load references. Not an assembled actuator drag or mass specification.

[ROLLVIS-CATALOG] Satellite Roller Screws — Catalogue 2018

Rollvis SA · Manufacturer PDF; printed pp. 2, 6, 11, 14–15 (PDF pp. 4, 8, 13, 16–17) · 2018 · Checked 2026-11-08

Describes lead versus pitch, direct/indirect efficiency, fatigue life, and driving torque. Use the selected nut configuration; no universal efficiency or ball-screw load multiplier is inferred.

[SCHAEFFLER-PWG] PWG Planetary Screw Drive — OPW Brochure

Schaeffler Technologies AG & Co. KG · Manufacturer brochure, September 2017, p. 1 · 2017 · Checked 2026-11-08

Family-level small-lead and load-capacity information. Does not establish PWG 10×1 ratings, assembly mass, or a measured breakaway force.

[THOMSON-LEAD-SCREWS] Lead Screws — Thomson Neff Precision Lead Screws & Supernuts

Thomson · Manufacturer brochure, printed p. 5 (PDF p. 5) · undated · Checked 2026-11-08

Self-locking depends on lead and friction; sufficiently high-lead sliding screws can backdrive. Includes torque and efficiency equations and critical-speed/support considerations.

Build a measurable actuator specification

Bring your torque-speed duty cycle, drag limits, supply and regeneration requirements, and mounting envelope. Use the acceptance plan above to turn an initial shortlist into testable joint requirements.