Motion control - frameless torque motors for humanoids

Executive summary

A frameless torque motor is a brushless permanent-magnet motor shipped as a separated stator and rotor, without a bearing, shaft, or housing. The integrator mounts the stator and rotor directly into the robot joint, and the robot’s own bearings carry the load. For humanoid robots, this is attractive because it removes duplicate mechanical structure, improves torque/weight ratio, shortens the joint stack, and lets the actuator share bearings with the robot structure.

The evidence tier here is important:

• A granted patent is legal evidence of recorded claims, not proof of thermal, torque-density, or lifetime performance.
• A peer-reviewed paper is better evidence of measured performance, but the specific motor in the best available paper is a small outrunner motor, not necessarily a production humanoid frameless torque motor.
• Supplier web pages are Tier 2 self-reported and often quote “typical” or “up to” performance.
• SEC filings are stronger evidence of supply-chain and adoption risk than marketing language.

The strongest source-bound claims in this note are:

• There is a granted U.S. patent for a compact direct-drive electromagnetic actuator: US 10,998,789 B2 “Electromagnetic actuator,” assigned to Genesis Robotics and Motion Technologies LP, granted 2021-05-04 [1]. It is evidence of recorded IP claims in direct-drive robotics motors, not evidence that a specific humanoid motor meets a specific torque-density number.
• A peer-reviewed IEEE/ASME Transactions on Robotics paper — DOI 10.1109/TRO.2016.2640183 — documents actuator design for dynamic legged robots and reports a candidate high-torque-density motor with 0.40 kg mass and 1.8 Nm peak torque, giving a peak torque density of 4.5 Nm/kg by calculation [2]. That paper is about legged-robot proprioceptive actuators, not a humanoid production guide, but it is the closest strict-tier evidence for motor torque-density selection in dynamic robots.

What a frameless torque motor is, and why humanoids care

A frameless torque motor is not a complete motor product. It consists of:

• a stator: generally laminated iron core with copper windings, often potted for thermal and mechanical integrity;
• a rotor: an array of permanent magnets, often air-gap or surface-mounted, with a hub return ring.

The customer provides the shaft, bearings, encoder/commutation sensor, and mechanical mounting. Product lines commonly cited by suppliers as frameless direct-drive motors include Kollmorgen’s TBM/TBM2G families [3], Celera Motion’s Omni and Aura frameless motors [4], TQ-RoboDrive’s frameless ILM drives [5], and Maxon frameless kits [6]. These are company product pages, so they establish that the product category exists and is sold; they do not independently prove any specific humanoid production deployment.

For a humanoid, the mechanical design logic is as follows:

• Fewer duplicate bearings and shafts reduce mass.
• Direct mounting shortens the joint stack, improving leg packaging.
• High pole-count, high-torque-density designs can reduce the need for large gear reductions, but humanoids usually still use gear reduction because direct-drive joint torque is rarely sufficient at biological joint volumes.
• Thermal integration with the robot link can improve continuous torque capability because the robot structure acts as heat sink.

This note tags marketing-level torque-density and cogging claims as self-reported unless separately cross-verified.

Source and methodology note

The usable strict evidence here is:

A patent is evidence of “we recorded a claimed invention.” It is not proof that the motor has production yield, thermal robustness, low cogging, or customer adoption. A paper is evidence of measured or manufacturer-specified values in a specific test context, but it is not evidence of commercial viability.

Any load-bearing technical claim from only one lineage is labeled [SINGLE-SOURCE]. For example, the U8 Pro motor mass and peak torque figures used below are from one peer-reviewed paper and are therefore marked single-source for the purpose of this note.

Patent landscape: what the granted patent does and does not show

Granted patent cited:

• US 10,998,789 B2 — “Electromagnetic actuator,” granted 2021-05-04, assignee Genesis Robotics and Motion Technologies LP [1].
• URL: patents.google.com
• Access date: 2026-04-26

The patent is in the direct-drive robotics-motor area. Genesis Robotics was later connected to Nidec’s direct-drive actuator efforts, which makes the patent relevant to humanoid joint motors, but the note does not treat a patent grant as independent performance evidence [1].

Moat read-through:

• A granted patent can block a specific electromagnetic architecture or manufacturing method.
• It does not by itself prevent competitors from using different frameless motor topologies.
• Many humanoid actuator supply chains use conventional radial-flux frameless motors from Kollmorgen, Celera Motion, TQ-RoboDrive, Maxon, and others, not necessarily this patent’s architecture [3–6].
• Therefore the moat is narrow and architecture-specific, not category-wide.

Discrepancy marker: supplier pages commonly imply “drop-in integration” and “high torque density,” but the technical paper used here shows a real high-torque-density motor with only 4.5 Nm/kg peak before gearing [2]. If a supplier claims an unspecified “high torque density,” that number is marketing unless a test condition, continuous rating, thermal class, and mass are disclosed. This is a source-class discrepancy, not a specific falsehood.

Peer-reviewed evidence: motor selection in dynamic legged robots

The peer-reviewed paper used here is:

• Wensing, P.M., Wang, A., Kim, S., et al., “Proprioceptive Actuator Design in the MIT Cheetah: Impact Mitigation and High-Bandwidth Physical Interaction for Dynamic Legged Robots,” IEEE/ASME Transactions on Robotics, 2017.
• DOI: 10.1109/TRO.2016.2640183 [2]
• URL: doi.org
• Access date: 2026-04-26

This paper is relevant to humanoid frameless torque motors because it addresses the same core question: how to select and package a high-torque-density motor for a dynamic direct-drive or near-direct-drive joint. It reports a candidate outrunner motor with:

• mass = 0.40 kg
• peak torque = 1.8 Nm
• torque constant = 0.106 Nm/A

These figures are reported here as single-source peer-reviewed values because they are not independently re-measured in this note [2].

5.1 Key engineering implication

The paper supports the central design tradeoff: for a dynamic legged robot, torque density and backdrivability matter more than raw electrical efficiency, because the actuator must deliver leg dynamics and impact absorption. Frameless/direct-drive motors reduce reflected inertia and improve force control, but peak torque density still forces the designer toward gearing.

Company technical documents and SEC filings

6.1 Company technical documents

The following supplier product pages establish that frameless torque motors are a commercial product category sold into robotics, medical, aerospace, and industrial motion:

• Kollmorgen frameless motor family: kollmorgen.com [3]
• Celera Motion frameless motors: celera.com [4]
• TQ-RoboDrive frameless drives: tq-robodrive.com [5]
• Maxon frameless motors: maxongroup.com [6]

These are Tier 2 self-reported. They are not independent benchmarks.

6.2 SEC filings and risk factors

SEC filing indexes are used here as primary-source corporate disclosure:

• Tesla, Inc. SEC EDGAR 10-K index: sec.gov [9]
• Allient Inc. SEC EDGAR filing index: sec.gov [10]
• Novanta Inc. SEC EDGAR filing index: sec.gov [11]
• Nidec Corporation SEC EDGAR filing index: sec.gov [12]
• Regal Rexnord Corporation SEC EDGAR filing index: sec.gov [13]
• Parker Hannifin SEC EDGAR filing index: sec.gov [14]
• Yaskawa Electric SEC EDGAR filing index: sec.gov [15]
• Moog Inc. SEC EDGAR filing index: sec.gov [16]

Cross-reference finding: supplier marketing emphasizes torque density and compactness, while corporate filings generally emphasize rare-earth magnet supply risk, customer concentration, competition, and inventory/demand volatility. For example, Tesla’s annual filing describes Optimus as development-stage and lists supply-chain and component-availability risks, which is more conservative than consumer-facing robot demos would imply [9]. This is a normal marketing-vs-filing asymmetry, not necessarily fraud, but it should prevent an investor from reading a motor product page as a humanoid production forecast.

No specific dollar value for frameless motor sales is used in this note because the public filings do not cleanly disaggregate frameless torque motors into a single line item. Any dollar figure would require primary-source extraction.

Quantitative worked calculation

This calculation starts from the peer-reviewed motor specification, because that is the strongest evidence in this note.

7.1 Peak torque-density calculation

Formula:

\[

\rho_{peak} = \frac{T_{peak}}{m_{motor}}

\]

Inputs from [2], marked single-source:

• \( T_{peak} = 1.8 \ \text{Nm} \)
• \( m_{motor} = 0.40 \ \text{kg} \)

Calculation:

\[

\rho_{peak} = \frac{1.8}{0.40} = 4.5 \ \text{Nm/kg}

\]

7.2 Sensitivity

These sensitivities assume mass and torque vary independently; in a real motor, changing torque usually changes mass because it changes copper/magnet volume.

7.3 Joint-level implication with gearing

Assume a humanoid knee joint target of \( T_{joint} = 80 \ \text{Nm} \). That value is assumed for illustration and is labeled needs primary source validation because it depends on robot mass, gait, and limb segment lengths. The calculation is:

\[

T_{motor, required} = \frac{T_{joint}}{N}

\]

with gear ratio \( N \). At \( N = 10 \):

\[

T_{motor, required} = \frac{80}{10} = 8 \ \text{Nm}

\]

Using the base peak torque density from [2]:

\[

m_{motor} = \frac{8}{4.5} \approx 1.78 \ \text{kg}

\]

This is close to the mass of a high-performance consumer motor, and it only satisfies peak joint torque, not continuous torque. Real humanoid knee actuators often use higher reduction, larger motors, or thermal management beyond what this simple calculation captures.

Interpretation: even with a well-selected direct-drive motor, the designer faces a peak torque-density ceiling; a gear reducer is the standard way to bridge the gap between biological joint torque and available motor torque. Frameless motors help mainly by reducing mass and packaging, not by eliminating gears.

Parameter table

Any row marked No or Single-source should not be treated as independent confirmation. In particular, supplier “torque density” claims without a test standard, continuous/peak definition, and thermal class are not comparable.

Unit economics / business model

Who monetizes: frameless motor suppliers monetize by selling the motor kit as a component, often with engineering support, custom winding, and integration services. Motors are usually sold per unit or per prototype/test quantity, with tooling or non-recurring engineering for custom variants [3][4][5][6].

How value is captured:

• The supplier captures component margin plus engineering service margin.
• The robot OEM/integrator captures the system margin by packaging the motor into a complete actuator with gearbox, encoder, brake, and drive.
• The magnet/copper/lamination suppliers capture material margin, especially rare-earth magnets.
• The bearing/encoder suppliers capture integration margin around the frameless kit.

Margins: no reliable frameless-motor-only gross margin is used here because public filings do not consistently disaggregate it. Motion-control segment margins at Novanta, Regal Rexnord, Parker, and Moog are available in SEC filings [11][13][14][16], but any specific margin applied to frameless torque motors must be extracted from primary filings and is currently needs primary source validation.

Unit economics logic: a frameless motor is cheaper to ship and lighter than a housed motor, but the customer must absorb design, assembly, thermal validation, and sensor alignment. That shifts cost from supplier to integrator and tends to reward suppliers who also sell design tools, thermal simulation, or integrated encoder/motor packages. For humanoids, Tesla’s public demonstration of custom rotary and linear actuators suggests the largest buyer may internalize actuator value rather than buy standard frameless motors at arm’s length [7][9].

Disconfirming Evidence

The thesis that frameless torque motors are a durable, high-value motion-control bottleneck for humanoids would be wrong if any of the following is true:

1. Humanoid production volumes remain negligible. If no humanoid OEM reaches high-volume production by 2027 or later, the addressable market for frameless motors stays small. Tesla’s filing already presents Optimus as development-stage [9].
2. Independent benchmarks fail to confirm torque-density advantage. If the only torque-density figures are supplier self-reports and independent test data show no meaningful advantage over housed motors, the economic case for integration complexity weakens.
3. Alternative architectures win. Axial-flux motors, integrated hub motors, or other direct-drive topologies could displace radial-flux frameless motors in the same joint slots.
4. Rare-earth supply and price risk materializes. Frameless torque motors depend on high-performance permanent magnets; a magnet supply disruption would raise cost or force design changes.
5. Patents do not block competition. The granted patent in this note is architecture-specific [1]. If competitors design around it and no broader patent wall exists, the “moat” is not durable.
6. Customers internalize too much value. If the largest humanoid OEMs, led by Tesla, continue to design and integrate custom actuators internally [7][9], independent motor suppliers could be reduced to low-margin component fabricators.