How Planetary Gearboxes Power Humanoid Robot Joints — And Why Two-Stage Designs Dominate

A full-size humanoid robot requires between 25 and 44 actuators depending on its degrees of freedom. Each one must generate meaningful torque, fit within the spatial envelope of a human limb, survive continuous dynamic loading, and remain responsive enough for the robot's control system to sense and react to external forces in real time. Across the industry, these actuators account for 40 to 60 percent of a robot's total bill of materials cost - making joint drivetrain selection one of the most consequential engineering decisions in the entire design process.

The difficulty is that a robot joint imposes demands that pull in opposite directions simultaneously:

High torque output is needed to support body weight and handle payloads - yet the gearbox must remain compact and lightweight enough to fit inside a limb.

Positional accuracy is needed for controlled manipulation - yet the transmission must also be backdrivable, allowing the joint to yield to external forces rather than resist them rigidly.

Structural stiffness is required to maintain control bandwidth and prevent limb oscillation under load - yet the drivetrain must also be compliant enough to absorb impact loads during walking and falling without amplifying reflected inertia back to the motor in ways that damage components.

Standard industrial gearboxes are often not suitable for direct application in humanoid robot joints due to design issues. Therefore, the two popular choices for robot joints today are harmonic and planetary drive, with planetary gearboxes gradually becoming more popular - especially two-stage planetary gearboxes.

A planetary gearbox gets its name from the way it moves: planet gears orbit a central sun gear, all contained within an outer ring gear.

Sun gear - connected to the motor shaft, this is the high-speed input.

Planet gears - typically three, they mesh simultaneously with both the sun gear and the ring gear, orbiting as they rotate.

Planet carrier - supports the planetary gears and rotates with their revolution, serving as the low-speed, high-torque output end.

Ring gear - the outer ring with internal teeth; in robot joint applications, it is typically held stationary to serve as the reaction housing.

The structural advantage is load distribution. Because multiple planet gears engage the ring gear simultaneously, the transmitted torque is shared across several contact points at once - not concentrated on a single gear mesh as in a parallel-shaft arrangement. The result is a significantly higher torque density within the same physical envelope.

For industrial machinery - conveyors, winches, mixing equipment - this is useful but not critical. The gearbox has room, weight is manageable, and back-driving is rarely a requirement.

Robot joints operate under entirely different constraints. The actuator must fit inside a human-scale limb, output meaningful torque from a compact motor, and remain transparent enough to the control system that force feedback is possible. A planetary gearbox, with its coaxial layout, high power-to-weight ratio, and efficiency above 96% per stage, maps naturally onto these requirements in a way that parallel-shaft gearboxes or worm drives do not.

This is also one of the reasons why planetary gearboxes have become the basic transmission architecture for robot joint actuators.

Three transmission types dominate humanoid robot joint design today: planetary gearboxes, harmonic drives, and RV reducers. Each has a distinct performance profile. Understanding where each excels - and where each falls short - is the foundation of any serious robot joint actuator selection process.

The Case Against Harmonic Drives in Dynamic Joints

Harmonic drives built their reputation in industrial robotic arms - stationary, precision-positioning tasks where zero backlash and high gear ratio in a compact package were the priority. For those applications, the trade-offs are acceptable.

Humanoid robot joints operate under fundamentally different conditions. Two limitations become critical:

Efficiency loss at scale

A humanoid robot gearbox isn't one unit - it's 25 to 44 units running simultaneously. At 70–85% efficiency per joint, the aggregate thermal and power losses across the full drivetrain are substantial, directly impacting battery runtime and thermal management.

Poor backdrivability

This is the more consequential issue. Backdrivability refers to how easily an external force can move a joint backward through the transmission. A robot joint actuator with low backdrivability is effectively "numb" - the control system cannot detect ground reaction forces, human contact, or unexpected loads below the threshold needed to overcome internal friction. For a legged robot, this makes stable walking and safe human interaction extremely difficult to achieve. Harmonic drives, due to their inherently higher internal friction, perform poorly here.

The flexspline - the thin, elastically deforming component at the heart of every harmonic drive - also introduces a fatigue life constraint under high-frequency impact loading, which is unavoidable in legged locomotion.

Where Each Type Belongs on a Humanoid Robot

In practice, most advanced humanoid robots today do not use a single transmission type throughout. The pattern that has emerged across the industry is broadly consistent:

Lower-body joints (hip, knee, ankle): High torque output, continuous impact loading, and the need for force transparency make this the natural domain of the planetary gearbox.

Upper-body joints (shoulder, elbow): Moderate torque with higher precision requirements. Planetary gearboxes remain competitive here, particularly in two-stage configurations where backlash can be held to 3 arcmin or below.

Distal joints (wrist, finger mechanisms): Minimal space, low torque, highest precision demands. Harmonic drives or low-backlash planetary variants are typically used where sub-arcminute positioning is required.

This joint-by-joint differentiation is why suppliers capable of producing planetary gearboxes across a range of backlash grades - rather than a single standard specification - are better positioned to support a complete humanoid robot joint system.

 

 

 

 

Single-stage planetary gearboxes work well in quasi-direct-drive (QDD) applications where a low gear ratio - typically 6:1 to 12:1 - is intentional. The goal there is maximum backdrivability and control transparency, at the cost of requiring a larger, heavier motor to generate useful output torque.

Most humanoid robot joints, however, need gear ratios well beyond what a single stage can deliver efficiently. This is where two-stage planetary gearboxes have become the standard answer.

The Ratio Gap a Single Stage Cannot Fill

A single planetary stage typically delivers ratios between 3:1 and 15:1. The joint-level torque requirements for hip and knee actuators - often in the range of 80 to 200 Nm output from motors spinning at 3,000 to 8,000 rpm - demand ratios of 20:1 to 80:1 to be met with a motor of practical size and weight.

Two stages in series multiply the ratio of each stage: a first stage at 6:1 combined with a second at 8:1 yields 48:1 overall. Critically, this is achieved while keeping the assembly coaxial and the axial length increase modest - typically 30 to 50% longer than a comparable single-stage unit, which remains acceptable within humanoid limb geometry.

What Two Stages Offer Beyond Ratio

Backlash distribution. In a two-stage planetary gearbox, positional error does not double because the backlash of the first (input) stage is divided by the reduction ratio of the second (output) stage. As a result, the total system backlash is heavily dominated by the output stage alone. With careful gear-mesh phasing and preload design, total system backlash can be held to 3 to 8 arcmin - within the acceptable range for most humanoid robot joint actuator applications outside the wrist.

Torque and shock capacity. Because a two-stage planetary gearbox allocates the massive final output torque entirely to the sturdier gears and planet carrier of the second stage, its mechanical life and structural strength far exceed those of a single-stage, high-reduction alternative of the same overall ratio when handling transient shock loads during walking and foot strike.

Industry Validation

The two-stage planetary configuration is not a theoretical preference - it reflects observable design choices across the industry.

Schaeffler's planetary gear actuator, unveiled at CES 2026 and developed specifically for humanoid robot joints, integrates a two-stage planetary gearbox with the motor, encoder, and controller in a single compact unit. The torque range cited - 60 to 250 Nm - is representative of what two-stage planetary designs can deliver at the joint level without resorting to heavier or less backdrivable alternatives.

From the custom manufacturing side, the pattern is equally consistent. The majority of joint actuator drawings received from humanoid robot developers in the UK, US, and Canada specify two-stage planetary configurations, with overall ratios typically falling between 25:1 and 60:1 and output torque targets that align with lower-body and mid-body joint requirements.

Standard catalogue gearboxes are built around common industrial ratios and interfaces. Humanoid robot joint actuators rarely fit them. Motor flanges, output interfaces, axial length budgets, and backlash grades are almost always joint-specific - which means the gearbox needs to be manufactured to drawing.

Gear Manufacturing Precision

Backlash and efficiency are determined at the gear-cutting and grinding stage. A supplier quoting P0-grade performance (≤ 1 arcmin) without ground gears at ISO/DIN Grade 5 is making a claim the process cannot support.

Three capabilities matter specifically for planetary robot joint gearboxes:

• Gear skiving - the most effective process for internal ring gears, producing consistent involute profiles on internal teeth
• Gear grinding - non-negotiable for Grade 5 accuracy and the surface finish quality that low-backlash mesh requires
• Fine-pitch range - planet gears for humanoid joints typically fall in the 0.5 – 2.0 module range; the supplier's equipment must cover this comfortably

Machining Tolerances for Housing and Carrier Components

Bearing seat tolerances directly affect preload, which determines both backlash and efficiency in the assembled gearbox. For P1 or P0 targets, critical bores and carrier shaft concentricity need to be held at ±0.005 mm - requiring precision grinding, not standard CNC milling. Non-critical structural features can be machined to ±0.01 mm without impacting performance.

From Drawing to Delivered Assembly

A capable supplier should cover the full process without the customer managing component sourcing separately:

Drawing review

Ratio feasibility, module selection, interface compatibility with the motor and output shaft, and any manufacturability concerns raised before tooling is committed.

Sample production

Backlash measurement, efficiency testing at rated torque, dimensional inspection, and material certification on a small initial batch.

Series production

Incorporating any adjustments from sample results before scaling.

Suppliers producing ring gear, planet gears, sun gear, carrier, and housing entirely in-house maintain tighter control over tolerance stack-up than those subcontracting individual components.

Three Problems That Appear Consistently

Planetary gearboxes - particularly two-stage configurations - have earned their position as the dominant transmission architecture for humanoid robot joints. The reasons are mechanical, not circumstantial: the combination of high torque density, acceptable backlash, genuine backdrivability, and competitive weight simply fits the joint actuator brief better than the alternatives across most of the robot's degrees of freedom.

Getting that performance in a custom joint gearbox comes down to two things: specifying the right parameters for each joint's actual load case, and working with a manufacturer whose gear production and machining capabilities match the precision grades the specification requires.

If you are developing a humanoid robot joint actuator and working from your own drawings - whether for individual planetary gear components or complete gearbox assemblies - we are happy to review your specifications and provide direct feedback on manufacturability, achievable backlash grade, and ratio feasibility.