Frameless Torque Motor — QDD Architecture & Design Tradeoffs

Method and source compliance

This note uses the requested tiering:

• Tier 1 technical/performance/IP claims are anchored where possible to a granted patent and peer-reviewed DOI-verified papers.
• Tier 2 adoption/application claims use manufacturer engineering documents and company filings, explicitly labeled as self-reported.
• No SEO content farms, anonymous benchmark sites, or AI-generated summaries are used.

A key source-bound caveat: a granted patent is evidence of a recorded invention, not proof of torque density, thermal capability, or commercial viability. The deepest evidence here for QDD feasibility is the peer-reviewed actuator work from Wensing et al. and Seok et al.; the patent cited is a granted series-elastic-actuator patent used for architecture contrast, not a QDD patent.

What is a frameless torque motor?

A frameless torque motor is a permanent-magnet synchronous motor sold as a rotor/stator kit without the housing, shaft, bearings, or end bells that an off-the-shelf housed motor includes. The integrator mounts the stator into the robot joint housing and the rotor onto the user’s shaft/hub. Kollmorgen’s frameless product pages describe the same category: frameless torque motors are intended for direct integration into a mechanism to reduce duplicate structure and improve torque-to-volume ratio (Kollmorgen TBM2G product page, tier 2, self-reported, accessed 2025-06-19).

The architecture is important for robot joints because integration cost is lower when the motor is not double-housed inside an already-housed actuator. But it also imposes a real engineering burden: air-gap alignment, rotor/stator concentricity, bearing stiffness, and thermal path must be designed by the integrator, not inherited from a motor vendor’s housing.

A frameless QR/QDD actuator typically includes:

1. Stator — laminated core with three-phase winding.
2. Rotor — permanent magnets on a back-iron ring, mounted to the output or to a sun-gear input.
3. Gear stage — in QDD, often a single-stage planetary or cycloidal reducer with a low ratio, typically 4:1 to 10:1.
4. Position feedback — absolute or incremental encoder, often on the motor side or output side.
5. Torque sensing — usually not a separate strain-gauge torque sensor; in QDD, torque is often inferred from motor current using the motor torque constant \(K_t\), though Wensing et al. also discuss direct torque sensing for impact mitigation (Wensing et al., IEEE T-RO 2017, DOI: 10.1109/TRO.2016.2640183, published 2017-06, accessed 2025-06-19).
6. Thermal path — stator conduction into the joint housing, sometimes with potting or a thermal pad.

Because no housing contains the rotor/stator, the frame itself is not “frameless” in terms of support; the robot joint structure becomes the motor frame.

What is QDD?

Quasi-direct-drive, or QDD, is an actuator topology that sits between a pure direct-drive motor and a high-reduction geared actuator. The term “quasi” refers to the low reduction ratio: enough gear reduction to amplify torque, but not enough to destroy backdrivability via high reflected inertia and high friction.

Wensing et al. describe the actuator approach in the MIT Cheetah as a “proprioceptive actuator” rather than QDD, but the architecture is the same: a high-torque-density electric motor, low gear reduction, integrated sensing, and control that relies on low mechanical impedance rather than series compliance (Wensing et al., IEEE T-RO 2017, DOI: 10.1109/TRO.2016.2640183, accessed 2025-06-19). The companion MIT Cheetah design papers also argue that actuator torque-density and regeneration are first-order robot design variables, not afterthoughts (Seok et al., IEEE/ASME TMech 2015, DOI: 10.1109/TMECH.2014.2339013, published 2015-06, accessed 2025-06-19; Seok et al., ICRA 2013, DOI: 10.1109/ICRA.2013.6631038, published 2013-05, accessed 2025-06-19).

The architectural claim is: a high-performance torque-dense motor plus a low-ratio gear gives near-direct-drive transparency while materially reducing motor mass. Pure direct drive requires the motor alone to provide full torque, which forces a large diameter and/or high current machine. A high-ratio gear makes the output non-backdrivable and can hide motor torque information behind friction and reflected inertia. QDD targets the middle.

QDD architecture deep dive

3.1 Mechanical architecture

A canonical QDD joint is a frameless torque motor mounted concentrically around the joint’s output shaft or sun gear. The rotor often rides on the same bearings as the joint itself, saving one bearing pair. A single-stage planetary gearset is a common choice because it has lower friction and lower reflected inertia than a strain-wave gear, at the cost of lower ratio.

Design variables include:

• Rotor inertia \(J_m\)
• Gear ratio \(N\)
• Gear efficiency \(\eta_g\)
• Gear friction torque \(\tau_f\)
• Motor torque constant \(K_t\) [N·m/A]
• Phase resistance \(R\) [Ω]
• Phase inductance \(L\) [H]
• Thermal resistance from winding to housing \(R_{th}\) [K/W]

The output torque and reflected inertia are the central trade:

\[

\tau_{out} = N \eta_g K_t I - \tau_f

\]

\[

J_{ref} = N^2 J_m + J_g

\]

where \(J_g\) is the reflected or output-referred gear inertia.

The dynamic equation for the output load \(J_l\) is approximately:

\[

\alpha_{out} = \frac{N \eta_g K_t I - \tau_f - \tau_{load}}{J_l + N^2 J_m + J_g}

\]

This equation makes the QDD trade explicit: gear ratio amplifies torque, but \(N^2\) amplifies rotor inertia. At N=10, a rotor’s inertia is multiplied 100× at the output; this can dominate acceleration if the load is small.

3.2 Electrical architecture

QDD actuators are almost always three-phase permanent-magnet synchronous motors driven by field-oriented control (FOC). Torque is commanded by a current loop. The electrical time constant of the motor sets a practical current-loop bandwidth ceiling:

\[

\tau_e = \frac{L}{R}

\]

and the approximate first-order bandwidth is:

\[

f_e \approx \frac{1}{2\pi \tau_e}

\]

This matters because the motor’s torque output depends on current; a low-inductance motor can change torque quickly, but may ripple and stress the inverter. QDD designs typically need a high current-loop bandwidth to compensate for impact and contact forces (Wensing et al., IEEE T-RO 2017, accessed 2025-06-19).

3.3 Thermal architecture

Frameless QDD motors can have high electromagnetic torque density, but they are generally thermally limited. Copper loss scales with current squared:

\[

P_{cu} = I^2 R

\]

and the winding temperature rise depends on thermal resistance:

\[

\Delta T = P_{cu} R_{th}

\]

The continuous torque rating is therefore set by how well the stator is heat-sunk into the joint housing. A frameless motor can have better thermal path than a housed motor because the stator can be potted or press-fit into the robot structure, but this is a design burden.

3.4 Control architecture

QDD actuators are intended to run in torque control, not just velocity or position control. The low reflected inertia means that contact forces are visible in motor current with less mechanical filtering. Wensing et al. use proprioceptive feedback—joint position, torque, and motor current—to implement dynamic locomotion and impact mitigation (Wensing et al., IEEE T-RO 2017, accessed 2025-06-19).

This differs from series elastic actuators, which deliberately insert a spring between gear output and load. The patent US 5,650,704 A is a granted SEA patent assigned to MIT; it describes an actuator that uses a compliant element to generate precise force control (US Patent 5,650,704 A, granted 1997-07-22, accessed 2025-06-19). The QDD design makes the opposite choice: remove series compliance and instead make the transmission and rotor low enough that force can be controlled directly from current and joint motion. Both patents and papers show that the field has explored both branches, but the QDD branch is the one most associated with low-inertia, high-bandwidth dynamic robots (Wensing et al., IEEE T-RO 2017, accessed 2025-06-19).

Design tradeoffs

4.1 Gear ratio and reflected inertia

The most important QDD tradeoff is the square-law reflected inertia term.

For a fixed rotor inertia:

If the load inertia is small, high \(N\) degrades acceleration more than it helps torque. QDD typically caps \(N\) around 10 because above that, reflected inertia and gear friction start to resemble a conventional geared actuator, even before considering friction.

4.2 Torque density vs speed

For a PMSM of fixed electrical power, torque constant and speed constant are inversely related. In SI units, for an ideal motor, \(K_t\) [N·m/A] and \(K_e\) [V/(rad/s)] are equal:

\[

K_t = K_e

\]

so a motor wound for high torque per amp inevitably has a low speed constant. A low gear ratio preserves output speed from this high-\(K_t\) motor while amplifying its torque. This is why QDD favors a high-\(K_t\), lower-speed frameless torque motor rather than a high-speed hobby motor with a large gearbox.

4.3 Backdrivability vs gear friction

Backdrivability is the ability of an external load to move the joint when the motor is unpowered or lightly driven. For QDD, backdrivability is not only about reflected inertia; it is also about static friction in the gear stage.

A single-stage planetary gear may have lower friction than a high-ratio strain-wave gear, but friction is still load-dependent and varies with manufacturing tolerance. If gear friction is too high, the actuator cannot be used as a torque-controlled impedance source at low torque amplitudes. The Wensing et al. design closely integrates a custom planetary gear with the motor to reduce this friction (Wensing et al., IEEE T-RO 2017, accessed 2025-06-19).

4.4 Thermal vs continuous torque

A frameless QDD motor may have high peak torque, but continuous torque is limited by thermal resistance. High \(K_t\) also means high torque per amp, but \(P_{cu}\) rises with \(I^2\). If the joint housing is small and has no active cooling, the continuous current—and therefore continuous torque—must be derated. This is a first-order design risk for humanoid or legged robots with tightly packed joints.

4.5 Compliance: QDD vs SEA

The granted SEA patent describes an actuator where a compliant element decouples the motor/gear inertia from the output to provide high-fidelity force control (US Patent 5,650,704 A, granted 1997-07-22, accessed 2025-06-19). SEA can be controllably soft and robust to impact, but the spring also limits closed-loop bandwidth and can add volume. QDD avoids the spring and instead relies on low mechanical impedance and current-based torque estimation. The design tradeoff is therefore:

• SEA: better force fidelity at low frequency and robustness, but lower mechanical bandwidth and more parts.
• QDD: higher bandwidth and fewer parts, but torque accuracy is finite because current-based torque estimation includes motor cogging, gear friction, and thermal drift unless external torque sensing is added (Wensing et al., IEEE T-RO 2017, accessed 2025-06-19).

4.6 Cost and lifetime

A low-ratio planetary gear is usually less expensive than a strain-wave gear, but QDD motor cost is higher per unit torque than a high-speed micro-motor because the frameless torque motor is larger and uses more magnet/copper volume. Lifetime is dominated by gear wear and bearing life. Direct-drive has no gear wear but adds motor mass and inverter cost. There is no public, third-party life-cycle comparison that resolves this for QDD as of 2025-06-19; vendor life ratings must be treated as self-reported.

4.7 Integration complexity

Frameless motors require tight runout and gap control. The robot manufacturer must hold bearing concentricity and stiffness; otherwise, air-gap variation causes torque ripple, cogging, or rotor-stator contact. Housed motors avoid this by controlling alignment in the motor factory, but then add duplicate structure and weight. This is a real manufacturing tradeoff, not a marketing detail, and it appears in vendor integration guides (Kollmorgen TBM2G product page, self-reported, accessed 2025-06-19).

Worked calculation

This worked calculation uses assumed representative values for a small humanoid/quadruped joint. The numbers are not sourced from a specific product datasheet and must be replaced with measured values before use in real design.

Inputs

These are assumed inputs, not measured values from a specific QDD actuator.

Motor continuous torque

\[

\tau_m = K_t I_c = 0.18 \times 5 = 0.90 \text{ N·m}

\]

Output continuous torque

\[

\tau_{out} = N \eta_g K_t I_c = 6 \times 0.85 \times 0.90 = 4.59 \text{ N·m}

\]

Reflected inertia

\[

J_{ref} = N^2 J_m = 6^2 \times 1.2 \times 10^{-5} = 4.32 \times 10^{-4} \text{ kg·m²}

\]

Per-phase copper loss

\[

P_{cu} = I_c^2 R = 5^2 \times 0.25 = 6.25 \text{ W}

\]

Electrical time constant and bandwidth

\[

\tau_e = \frac{L}{R} = \frac{0.15 \times 10^{-3}}{0.25} = 0.0006 \text{ s}

\]

\[

f_e = \frac{1}{2\pi \tau_e} = \frac{1}{2\pi \times 0.0006} \approx 265 \text{ Hz}

\]

Thermal rise

\[

\Delta T = P_{cu} R_{th} = 6.25 \times 1.5 = 9.4 \text{ K}

\]

Sensitivity check

• If \(N\) rises from 6 to 9, output torque at same current rises by \(9/6 = 1.5\)× to about 6.89 N·m, but reflected inertia rises by \(81/36 = 2.25\)× to \(9.72 \times 10^{-4}\) kg·m².
• If \(K_t\) rises from 0.18 to 0.25 N·m/A, output torque rises by \(0.25/0.18 = 1.39\)× to about 6.38 N·m, but for the same winding volume the speed constant falls proportionally.
• If \(R\) rises from 0.25 to 0.30 Ω, per-phase copper loss at 5 A rises from 6.25 W to 7.50 W; thermal rise at \(R_{th}=1.5\) K/W rises from 9.4 K to 11.25 K.
• If \(N\) falls to 4, reflected inertia falls to \(1.92 \times 10^{-4}\) kg·m², but output torque falls to 3.06 N·m at the same current.

Conclusion of calculation: For this representative small joint, N=6 is a plausible QDD compromise: output torque is amplified 6×, while reflected inertia remains under \(5 \times 10^{-4}\) kg·m². The dominant sensitivities are gear ratio and thermal resistance, not torque constant alone.

Parameter table

The table intentionally distinguishes assumed worked-example inputs from sourced architecture claims. No vendor torque-density number is used as an independent factual claim.

Unit economics / business model

The frameless-torque-motor QDD value chain monetizes at several layers:

1. Motor kit producers — Kollmorgen, Celera Motion/Novanta, TQ-RoboDrive, Allient, Parker, Nidec, Yaskawa. They monetize by selling rotor/stator kits, engineering time, and matched drives. This is a high-mix, B2B component business; specific QDD gross margins are not disclosed in SEC filings. Fortive’s 10-K reports segment-level results but does not isolate Kollmorgen frameless motor margins (Fortive SEC EDGAR page, accessed 2025-06-19). Any QDD-specific margin number would be “needs primary source validation.”

1. Actuator integrators — robot OEMs and actuator module makers. They monetize the final joint or full robot. Value capture here is in integration, control software, and application-specific mechanical design, not in the raw motor kit alone.

1. Drive and encoder suppliers — inverter vendors and encoder vendors capture incremental content because QDD requires high-bandwidth FOC and good position feedback.

The unit economics are therefore not a royalty or subscription model; they are component sales plus engineering services. No recurring revenue stream from the motor itself is visible except replacement gear stages, spare stators/rotors, and service.

The key unresolved unit-economics question is whether QDD’s higher motor cost and integration cost are offset by deleting the strain-wave gear or series-elastic spring and by simplifying torque sensing. Public filings do not yet provide a clean answer.

Disconfirming Evidence

The QDD thesis would be wrong or materially weakened if:

1. A peer-reviewed or manufacturer-independent teardown shows that QDD backdrivability at target torque levels is not materially better than a conventional housed brushless motor with a low-ratio gearbox at the same cost and mass.
2. Robot OEMs shift toward high-reduction series elastic actuators because QDD’s current-based torque estimation cannot meet accuracy, safety, or thermal requirements; this would reverse the design trend implied by the Wensing et al. actuator line (Wensing et al., IEEE T-RO 2017, accessed 2025-06-19).
3. Frameless motor integration costs—bearing alignment, air-gap control, thermal potting, and custom drives—make QDD cost-per-N·m higher than alternative architectures, and no public company filing shows QDD-specific revenue growth.
4. A granted QDD-specific patent, if granted, is cited as evidence of performance without independent measurement; this note treats granted patents as claims only, not proof.