Actuator

An actuator is the part of a machine that converts an energy input, electric current, pressurized hydraulic fluid, or compressed air, into controlled mechanical motion. The word comes from the Latin actuare, "to put into action," by way of Medieval Latin actuatus^[1]. The term applies across all of engineering, but on AI Wiki it refers almost exclusively to the joint-level modules that move a humanoid robot or other powered robot. In that narrower sense, a modern robotic actuator packages a motor, a gearbox or screw, a position sensor, and control electronics into one self-contained unit that bolts directly into a robot's structure and drives a single joint^[2][3]. Industry analysts have identified actuators as the largest single cost category in a humanoid's bill of materials, commonly estimated at 50 to 70 percent of total component cost^[4].

In brief: a useful comparison is a single muscle-and-joint unit rather than a whole limb. A human arm moves through dozens of separate muscles pulling on a shared skeleton. A robot arm instead uses one sealed, replaceable actuator per joint, with its own motor, gearing, and sensors built in. Engineers classify actuators by what powers them (electric, hydraulic, or pneumatic) and by how they move (linear or rotary), and those two choices largely determine a robot's cost, weight, precision, and safety around people.

How an actuator works

Every robotic actuator performs the same basic conversion: it turns an energy input into force or torque at an output shaft or rod, then holds or modulates that output under closed-loop control. In the electric design used across nearly all current humanoids, that conversion chain has four parts^[2][3]:

• A motor, typically a brushless DC motor or a closely related servo motor, spins at high speed but comparatively low torque.
• A reducer, a gearbox or screw mechanism, trades that speed for torque, the actuator's most consequential design choice, covered in detail below.
• An encoder, typically a rotary encoder, reports the shaft's exact position back to the controller thousands of times per second, closing the position-control loop.
• A force/torque sensor, often a strain-gauge element built into the output stage, measures how hard the joint is pushing or being pushed, letting the robot sense contact and control force rather than just position^[5].

These parts sit on bearings, commonly compact rolling-element bearings or, where radial space is tight, thin-section crossed roller bearings, and share a control board that runs motor commutation and talks to the robot's central computer. Integrating all of this into one housing, rather than assembling the parts separately, is what turns a generic motor into a purpose-built robot actuator: it shortens wiring, keeps sensors close to the load they measure, and lets a technician swap one failed joint without disassembling an entire limb^[3].

Types by power source

An actuator's power source sets a hard ceiling on what it can do well. Only three energy carriers are practical for driving robot motion: electric current, pressurized liquid, and compressed gas, each implying a different mechanical design and set of compromises^[6].

Electric

Electric actuators use a motor, directly for rotary output or through a mechanical converter for linear output, and they dominate humanoid robotics almost completely. The appeal is architectural as much as it is about raw performance: electricity distributes through simple copper wires instead of pressurized hoses, it is easy to meter and switch with solid-state electronics, and it integrates naturally with the position and force sensors a robot needs for closed-loop control^[6][7]. Brushless DC motors commonly reach 85 to 95 percent efficiency at the motor itself, well above brushed or induction alternatives^[8], though the complete actuator's efficiency is lower once gearbox or screw losses are added in (see below). The tradeoff is raw force density: a bare electric motor cannot match a hydraulic cylinder of the same size, which is why humanoid designers lean on high-ratio gearboxes and screws to multiply an electric motor's modest torque.

Hydraulic

Hydraulic actuators use a pump to pressurize an incompressible liquid, then release it into a cylinder to drive a piston. Because liquid barely compresses, hydraulic systems transmit force directly and can hold a heavy load steadily, which is why hydraulics dominate excavators, aircraft control surfaces, and other heavy machinery^[9]. A 2023 peer-reviewed comparison of electric, hydraulic, and pneumatic linear actuators performing an identical motion cycle found the electric system drawing 234 to 314 watts, while the hydraulic system drew roughly 650 watts in steady motion and over 1,500 watts during valve activation at high pressure, and its test rig weighed nearly five times as much as the electric one^[10]. That combination, a heavy pump, a network of hoses and valves, and fluid that must be sealed against leaks near people, has pushed most humanoid builders away from hydraulics for primary body joints. The clearest example is Boston Dynamics' original Atlas, hydraulically actuated for eleven years after its 2013 debut before the company retired it in 2024 for an all-electric successor^[11][12]. A few companies still use hydraulics selectively, usually in the hands, described under artificial-muscle alternatives below.

Pneumatic

Pneumatic actuators work like hydraulic ones but use compressed air instead of liquid. Air is cheap, clean, and safe to vent, and pneumatic cylinders can move very fast, but air is compressible, so it behaves like a spring: pneumatic actuators are hard to stop at a precise position or hold rigidly under load without extra valving and sensors^[6][13]. In the same 2023 comparison cited above, the pneumatic system drew 920 to 940 watts regardless of pressure setting, the highest and least efficient of the three^[10]. These traits make pneumatics a poor fit for a humanoid's primary walking and manipulation joints, but a reasonable one for simple binary motions (open or close, extend or retract) and for soft grippers, where natural compliance is an advantage rather than a flaw.

Types by motion: linear and rotary

Independent of power source, an actuator produces either linear motion (a straight-line push or pull) or rotary motion (rotation around an axis). Pneumatic and hydraulic actuators are naturally linear, since their core mechanism is a piston sliding in a cylinder, though both can drive rotary output through specialized designs. Electric actuators are naturally rotary, since a motor's native output is a spinning shaft; producing linear motion electrically requires an added mechanical stage to convert rotation into translation^[14].

Rotary joints

Most humanoid joints rotate, so most humanoid actuators are rotary, built from a motor plus a reduction gearbox. Three gearbox families, plus one low-ratio alternative, cover nearly all current humanoid designs.

A harmonic drive, also called a strain wave gear, uses a flexible splined cup that deforms as it is driven by an elliptical wave generator, engaging roughly 30 percent of its teeth at any instant. That broad simultaneous contact gives harmonic drives near-zero backlash, typically under one arc-minute, and among the best torque-to-weight ratios of any common gearbox, at the cost of moderate efficiency, roughly 70 to 90 percent depending on ratio and load, and a known failure mode: the flexible cup can develop fatigue cracks under the repeated shock loading of walking^[15][16][17]. Harmonic drives are the default choice for Tesla Optimus's shoulder and hip joints and for most collaborative-robot arms^[18].

A cycloidal drive uses an eccentric cam to orbit a lobed disc against a ring of fixed pins, transferring load across several rolling contact points instead of through a flexing part. That rolling contact makes cycloidal drives markedly more tolerant of shock and impact than harmonic drives, at somewhat higher single-stage efficiency, roughly 85 to 93 percent, but with more backlash, typically 3 to 5 arc-minutes unless preloaded, and a bulkier package for the same ratio^[16][19]. Many industrial robot joints use an "RV reducer," which pairs a first-stage planetary gear train with a second-stage cycloidal disc in one housing, a combination long associated with the Japanese supplier Nabtesco^[19][20].

A bare planetary gear train, several small gears orbiting a central sun gear inside a ring gear, is simpler and more efficient than either strain-wave or cycloidal designs at a given ratio, but has more backlash unless the gears are preloaded. In December 2025 the bearings supplier Schaeffler unveiled an integrated humanoid actuator built around a two-stage planetary gearbox, packaged with its own motor, encoder, and controller in one housing, spanning a torque range of 60 to 250 newton-meters. Schaeffler's own materials put the number of such actuators needed per humanoid at roughly 25 to 30, broadly consistent with counts reported elsewhere in the industry^[20].

A quasi-direct drive actuator, usually abbreviated QDD, takes the opposite approach: instead of a high gear ratio, it pairs an oversized, torque-dense motor with a low ratio, cited across different sources as anywhere from roughly 1:1 up to about 10:1, versus 30:1 or higher for harmonic and cycloidal designs. Engineer Ben Katz popularized the approach while building MIT's Mini Cheetah quadruped in the late 2010s, working in Sangbae Kim's lab^[21][22]. Because so little gearing sits between the motor and the joint, QDD actuators stay backdrivable (see below), respond quickly, and are mechanically simple, though they give up the raw torque density of a heavily geared harmonic or cycloidal design. Unitree has built its humanoid and quadruped lineup heavily around in-house QDD actuators, which one industry analysis described as up to 80 percent cheaper to produce than harmonic-drive alternatives of comparable capability^[4].

Linear joints

Where a joint needs straight-line force, electric actuators convert a motor's rotation into linear travel with a threaded mechanism. Two designs dominate humanoid use.

A ball screw uses a recirculating chain of steel balls rolling in helical grooves between a screw shaft and a nut, converting rotation to translation at relatively low friction, typically around 80 to 90 percent mechanical efficiency^[23]. Ball screws are the cheaper, more mature option and remain common in humanoid limb actuators, but the balls carry load at essentially a point contact, which limits how much repeated shock the screw can absorb before pitting or fatigue failure.

A planetary roller screw replaces the balls with several threaded rollers arranged around the shaft like planets around a sun, each in continuous line contact with the screw and the nut rather than point contact. That geometry gives a roller screw roughly three to five times the dynamic load rating, and up to about ten times the static load rating, of a same-size ball screw, with a correspondingly longer fatigue life, at efficiency that can reach the low-to-mid 90 percent range under favorable conditions^[24][25]. Tesla adopted roller screws over ball screws for Optimus's knee and ankle actuators specifically for that shock tolerance^[26], and trade press reports Boston Dynamics' new electric Atlas relies on them too^[27]. The catch is cost and supply: roller screws need tighter manufacturing tolerances than ball screws, only a handful of makers worldwide can produce them at humanoid-grade precision, and Wall Street analysts have priced individual units at $1,350 to $2,700, far above a comparable ball screw^[28][29]. A humanoid can use 40 or more screws, so screws alone can account for up to a third of total build cost, and a Morgan Stanley research note in February 2025 predicted that roller screws would eventually make up the majority of screws used in humanoids^[28][29].

A plain leadscrew, a threaded shaft turning in a matched nut with sliding rather than rolling contact, is cheaper and simpler than either but far less efficient and more prone to wear, so it appears mainly in cost-sensitive or low-cycle applications rather than primary humanoid joints.

Scroll sideways for more →

Key evaluation criteria

Choosing an actuator means trading off several properties at once, and no single power source or mechanism wins on all of them^[30].

Power density describes how much force or torque an actuator produces per unit of weight or volume. Hydraulic systems have historically led on raw power density, which is part of why they remain common in heavy machinery, but electric designs using compact high-pole-count motors and lightweight gearboxes have closed much of the gap at humanoid scale^[31].

Controllability describes how precisely position and force can be commanded and held. Electric actuators pair naturally with digital control electronics and win decisively here; pneumatic actuators lose the most ground, because compressed air's springiness makes it hard to stop and hold a position without extra sensing and valving^[6].

Backdrivability describes whether an external force can move a joint with relatively little resistance, rather than the joint being effectively locked by internal friction and gearing^[32]. A backdrivable joint lets a robot sense a bump or a person's touch through its own motor current, and lets the joint absorb impact instead of transmitting it into the structure. Low-ratio and quasi-direct-drive actuators are inherently more backdrivable than high-ratio harmonic or cycloidal designs, which is why QDD and related compliant designs remain popular for legs and other joints that must react to unexpected contact.

Efficiency describes how much input energy reaches the output as useful motion, which matters directly for a battery-powered robot's runtime. Electric systems win clearly; hydraulic and pneumatic systems lose energy continuously to maintaining line pressure, valve throttling, and, for pneumatics, the thermodynamic cost of compressing and expanding air^[10].

Response speed describes how quickly an actuator's output can change after a new command, often summarized as control bandwidth, which matters for tasks like catching a stumble. Pneumatic and hydraulic systems move fast once a valve opens, but the valve itself adds lag; low-inertia electric designs, especially QDD actuators, achieve high torque bandwidth because little gearing sits between command and output^[33].

System complexity describes how many parts, and what supporting infrastructure, an actuator needs. Electric actuators need only power and data wiring; hydraulics need a pump, valve manifold, reservoir, and a body full of hoses that can leak; pneumatics fall in between, needing a compressor and air lines but no fluid to contain^[6].

Scroll sideways for more →

Weighed together, these criteria explain why electric actuators became the default for humanoid robots. They do not lead on every single axis, hydraulics still win on raw power density, but they offer the strongest overall combination of control, efficiency, and simplicity for a battery-powered machine that has to work safely around people^[6][7].

Use in humanoid robots

A humanoid robot's actuators function as distributed muscles: dozens of separate, independently controlled units standing in for the hundreds of muscles in a human body. Tesla's Optimus is the most extensively documented example. The original "Bumblebee" prototype, shown at Tesla's AI Day in September 2022, used 28 structural actuators across the body, standardized down to just six distinct actuator types to simplify manufacturing, plus a separately engineered hand targeting 11 degrees of freedom^[34]. The Generation 2 Optimus revealed in December 2023 carried a similar body architecture forward, with rotary harmonic-drive actuators at the shoulders and hips and linear planetary-roller-screw actuators at the knees and ankles, chosen there specifically for shock tolerance during walking, and its hands reached that same target^[26][35].

The often-repeated "28 actuators" figure describes that body architecture, not the whole robot, and it has not stayed fixed. Tesla's Generation 3 hand and forearm assembly, entering low-volume production in 2026, jumped to 22 degrees of freedom per hand using a tendon-driven design that relocates actuators from the hand into the forearm, echoing how human forearm muscles pull finger tendons rather than sitting inside the fingers themselves. According to reporting on Tesla's related patent filings, CEO Elon Musk said on X in late 2025 that each arm now packages 25 actuators in its forearm and hand, 23 driving the hand and two the wrist, for 50 actuators across both hands alone, on top of the body's original complement^[36]. Hand actuation, in other words, has grown far faster across Optimus generations than the body-joint count that made "28 actuators" a widely repeated figure in the first place.

Other humanoid builders illustrate the same tradeoffs from different angles. Boston Dynamics ran its original Atlas on 28 hydraulically actuated degrees of freedom for eleven years, prized for power density in a compact frame, before retiring it in April 2024 for an all-electric successor^[11][12]. Trade press has attributed the new Atlas's force density to planetary-roller-screw linear actuators and high-density neodymium-magnet motors, chosen to match the old hydraulic system's output without the fluid maintenance, though Boston Dynamics itself has published few technical specifics^[27][37]. Unitree has taken nearly the opposite path from Tesla, building actuators in-house around low-cost QDD designs rather than high-precision harmonic drives, part of a vertical-integration strategy that one industry analysis credited with helping cut list prices for the Unitree G1 from more than $50,000 to roughly $27,000 within about a year, with some bulk deals reportedly well under $20,000^[4]. Figure AI has pursued a similar cost curve from the other direction: its custom actuator cost per unit reportedly fell from roughly $1,200 in the 2023 Figure 01 prototype to under $400 in the 2025 Figure 03, attributed to simplified gearing and a shift from CNC-machined parts to higher-volume die casting and injection molding^[38].

Across the industry, a working range of roughly 25 to 30 actuators for a humanoid's core body joints recurs often enough, in Tesla's original design and in Schaeffler's estimate, to serve as a reasonable rule of thumb, though exact counts vary by design philosophy and hand actuators are typically counted separately^[20][34].

Artificial-muscle alternatives

A small number of companies bypass the motor-plus-gearbox-plus-screw formula entirely, betting that fluid-driven or fiber-based artificial muscles can outperform conventional actuators on power density and compliance, at the cost of new engineering problems around valves, fluid routing, and sealing. This approach has a long history in soft robotics: pneumatic McKibben-style muscles, braided sleeves that contract when inflated, date to the 1950s and remain common in prosthetics and simple grippers^[39].

Sanctuary AI uses miniaturized hydraulic valves, described as coin-sized and driven by food-safe oil, to actuate the 21 degrees of freedom in its Phoenix humanoid's hands. The company says the approach delivers an order of magnitude higher power density than cable-driven or electromechanical hand designs, and that it has tested valve actuators past 2 billion cycles without leakage or measurable degradation^[40][41]. These are Sanctuary AI's own figures rather than independently audited results, but the underlying logic, that hydraulic actuation packs more force into a finger's tight volume than a motor and gearbox can, matches hydraulics' power-density advantage described above.

Clone Robotics, a Polish company, has taken the artificial-muscle concept furthest with its Myofiber technology: monolithic synthetic muscle fibers attached directly to a polymer skeleton modeled on human anatomy, which the company says contract in under 50 milliseconds, shorten by more than 30 percent unloaded, and generate at least a kilogram of force per 3-gram fiber^[42]. Clone's early Protoclone prototype, shown in 2024, used roughly 1,000 Myofiber muscles driven by pressurized oil or air through 3D-printed flexible hoses, powered by a 500-watt pump the company nicknamed an "artificial heart"^[43][44]. Its current production android, described on Clone's own site as running on "water and electricity," uses 910 muscle fibers to produce 164 degrees of freedom, having shifted toward water-based hydraulics^[42][45]. Both companies sidestep the gearbox and precision-screw costs that dominate conventional actuator bills of materials, trading them for different hard problems: valve manufacturing, leak-free sealing at high cycle counts, and continuous fluid management.

Suppliers and landscape

No single company sells a complete humanoid actuator off the shelf the way a car buyer might source an alternator. The supply chain instead splits between component specialists supplying motors, gearboxes, screws, and bearings, and the robot makers who integrate those parts, or design comparable ones in-house.

Scroll sideways for more →

Several leading humanoid developers, including Tesla, Figure AI, Unitree, Boston Dynamics, and 1X Technologies, design and build most of their actuators in-house rather than buy them, treating actuator design as core intellectual property.

See also

• Humanoid robot
• Degrees of freedom
• Dexterous hand
• Tendon-driven
• Tesla Optimus
• Boston Dynamics

References

1. Douglas Harper, "actuate," Etymonline. https://www.etymonline.com/word/actuate ↩
2. EYOU Robotics, "Types of Robot Joint Actuators: Harmonic, Planetary, QDD & More." https://eyoubot.com/en/blog/types-of-robot-joint-actuators ↩
3. iNetic Motion, "Integrated Robotic Joint Actuator Applications." https://ineticmotion.com/applications/robotic-joints/ ↩
4. SemiAnalysis, "Unitree's Impossible Trajectory Is Still Overlooked." https://newsletter.semianalysis.com/p/chinas-unitree-will-dominate-global ↩
5. FUTEK, "Humanoid Robot Sensors." https://www.futek.com/applications/humanoid-robot-sensors ↩
6. Progressive Automations, "Hydraulic vs Pneumatic vs Electric Actuators: Pros and Cons." https://www.progressiveautomations.com/blogs/products/pros-cons-of-hydraulic-pneumatic-and-electric-linear-actuators ↩
7. TCI Supply, "Pneumatic vs Hydraulic vs Electric Actuators: How to Choose?" https://tcisupply.com/pneumatic-vs-hydraulic-vs-electric-actuators-differences/ ↩
8. Assun Motor, "Understanding DC Brushless Motor Efficiency & How to Test For It." https://assunmotor.com/blog/dc-brushless-motor-efficiency/ ↩
9. Janhen Valve, "Actuators 101: Pneumatic vs. Electric vs. Hydraulic." https://janhenvalve.com/pneumatic-vs-electric-vs-hydraulic-actuators-comparison/ ↩
10. Scientific Reports (Nature), "Comparison of hydraulic, pneumatic and electric linear actuation systems" (2023). https://www.nature.com/articles/s41598-023-47602-x ↩
11. Wikipedia, "Atlas (robot)." https://en.wikipedia.org/wiki/Atlas_(robot) ↩
12. Boston Dynamics, "An Electric New Era for Atlas" (April 17, 2024). https://bostondynamics.com/blog/electric-new-era-for-atlas/ ↩
13. IEEE Spectrum, "Hello, Electric Atlas." https://spectrum.ieee.org/atlas-humanoid-robot ↩
14. RoboticsTomorrow, "Linear Actuators vs Rotary Actuators: The Core Choice for Humanoid Robot Joints." https://www.roboticstomorrow.com/article/2025/10/linear-actuators-vs-rotary-actuators-the-core-choice-for-humanoid-robot-joints/25703 ↩
15. PlaPivot, "Harmonic Drive vs Cycloidal Drive: Which Gearbox for Your Robot Joint." https://plapivot.com/blog/blog_harmonic-drive-vs-cycloidal ↩
16. Cone Drive, "Harmonic Drive Vs Cycloidal." https://conedrive.com/cycloidal-gears-vs-harmonic-gears/ ↩
17. Firgelli, "Humanoid Robot Actuators: The Engineering Reality Check." https://www.firgelli.com/blogs/news/the-physics-of-humanoid-motion ↩
18. CubeMars, "Boston Dynamics, Shift from Hydraulic to Electric Actuation: A New Era in Robotics." https://www.cubemars.com/boston-dynamics-hydraulic-electric-actuation.html ↩
19. How To Mechatronics, "Harmonic vs Cycloidal Drive: Torque, Backlash and Wear Test." https://howtomechatronics.com/how-it-works/harmonic-vs-cycloidal-drive-designing-3d-printing-testing/ ↩
20. Robotics and Automation News, "Schaeffler unveils innovative planetary gear actuator for humanoid robots" (December 30, 2025). https://roboticsandautomationnews.com/2025/12/30/schaeffler-presents-innovative-planetary-gear-actuator-for-humanoid-robots/97939/ ↩
21. Hackaday, "A Budget Quasi-Direct-Drive Motor Inspired By MIT's Mini Cheetah" (July 14, 2025). https://hackaday.com/2025/07/14/a-budget-quasi-direct-drive-motor-inpired-by-mits-mini-cheetah/ ↩
22. MDPI Proceedings, "Design of a Quasi-Direct-Drive Actuator for Dynamic Motions" (2020). https://www.mdpi.com/2504-3900/64/1/11 ↩
23. Machine Design, "What's the Difference Between Roller and Ball Screws?" https://www.machinedesign.com/mechanical-motion-systems/article/21175536/whats-the-difference-between-roller-and-ball-screws ↩
24. Tolomatic, "How roller-screw and ball-screw actuators compare in high-force applications." https://www.tolomatic.com/info-center/resource-details/how-roller-screw-and-ball-screw-actuators-compare-in-high-force-applications/ ↩
25. Moog, "Selection of Ball Screw versus Roller Screw Technologies." https://www.moog.com/news/ideas-in-motion-control/2004/08/selection-of-ball-screw-versus-roller-screw-technologies.html ↩
26. Wikipedia, "Optimus (robot)." https://en.wikipedia.org/wiki/Optimus_(robot) ↩
27. Brian D. Colwell, "A Complete Review Of Boston Dynamics' Atlas Robot." https://briandcolwell.com/a-complete-review-of-boston-dynamics-atlas-robot/ ↩
28. Fast Company, "This tiny screw is powering the humanoid robot revolution." https://www.fastcompany.com/91314612/this-tiny-screw-is-powering-the-humanoid-robot-revolution ↩
29. Holistic News, "Planetary Roller Screws: Essential for Robotics" (June 25, 2025). https://holistic.news/en/planetary-roller-screws-essential-for-robotics/ ↩
30. Emergent Mind, "Backdrivable Actuators in Robotics." https://www.emergentmind.com/topics/backdrivable-actuators ↩
31. CubeMars, "Selection Guide for Humanoid Robot Knee and Hip Joint Motors." https://www.cubemars.com/how-to-choose-hip-and-knee-joint-motors-for-humanoid-robots.html ↩
32. Enrique del Sol, "Backdrivability," Robotics, Mechatronics & Control Research. https://enriquedelsol.com/2017/12/05/backdrivability/ ↩
33. Eppinger and Seering, "Understanding Bandwidth Limitations in Robot Force Control," MIT/CMU. http://www.cs.cmu.edu/~cga/force-control/eppinger2.pdf ↩
34. James Morris, "Tesla AI Day 2022: Musk Demonstrates Optimus Humanoid Robot For Under $20,000," Forbes (October 1, 2022). https://www.forbes.com/sites/jamesmorris/2022/10/01/tesla-ai-day-2022-musk-promises-optimus-humanoid-robot-for-under-20000/ ↩
35. Inspire Robots, "A Technical Breakdown of Tesla Optimus' Linear Actuators," LinkedIn Pulse. https://www.linkedin.com/pulse/inspire-robots-technical-breakdown-tesla-optimus-linear--1c ↩
36. Electric-vehicles.com, "Tesla Files Patents Revealing Optimus Gen 3 Mechanical Blueprint." https://eletric-vehicles.com/tesla/tesla-files-patents-revealing-optimus-gen-3-mechanical-blueprint/ ↩
37. ROBOTS: Your Guide to the World of Robotics, "Atlas." https://robotsguide.com/robots/atlas ↩
38. Robotomated, "Figure AI Robot Review: Figure 03 Specs, Capabilities, and Real-World Performance." https://robotomated.com/learn/humanoid/figure-ai-robot-review ↩
39. MDPI Actuators, "Recent Developments in Pneumatic Artificial Muscle Actuators." https://www.mdpi.com/2076-0825/14/12/582 ↩
40. Sanctuary AI, "Sanctuary AI Demonstrates In-Hand Manipulation Capabilities for Improved General Purpose Robot Dexterity." https://www.sanctuary.ai/blog/sanctuary-ai-demonstrates-in-hand-manipulation-capabilities-for-improved-general-purpose-robot-dexterity ↩
41. Mike Kalil, "Back to Hydraulics? Clone and Sanctuary AI Boast Next-Gen Fluid-Powered Robots." https://mikekalil.com/blog/sanctuary-clone-hydraulics/ ↩
42. Clone Robotics, "Android." https://clonerobotics.com/android/ ↩
43. R&D World, "Protoclone V1 Is a Sweating Robot With 1000 Artificial Muscles." https://www.rdworldonline.com/protoclone-v1-1000-artificial-muscles-power-this-sweating-robots-human-like-moves/ ↩
44. Industry Insider, "Clone Alpha: A Humanoid Robot Built With Synthetic Organs and Artificial Muscles." https://industryinsider.eu/automation-and-production-lines/clone-alpha-a-humanoid-robot/ ↩
45. Interesting Engineering, "Water-Powered Humanoid Robot With Synthetic Organs, Muscles Unveiled." https://interestingengineering.com/innovation/clone-alpha-humanoid-robot-unveiled-poland ↩
46. Zinnia Lee, "Brothers Ride Humanoid Robot Wave To Become China's New Billionaires," Forbes (April 23, 2026). https://www.forbes.com/sites/zinnialee/2026/04/23/humanoid-mania-turns-chinese-brothers-behind-robotic-joint-maker-into-billionaires/ ↩
47. KrASIA, "Behind the Robotics Boom, ZeroErr Raises Funds to Build the Parts That Power It." https://kr-asia.com/behind-the-robotics-boom-zeroerr-raises-funds-to-build-the-parts-that-power-it ↩
48. RoboticsCenter Developer Wiki, "Damiao QDD Motors, Reference Guide." https://www.roboticscenter.ai/wiki/damiao-motors ↩
49. Linear Motion Tips, "SKF Motion Technologies Becomes Ewellix." https://www.linearmotiontips.com/skf-motion-technologies-becomes-ewellix/ ↩
50. Drives & Controls, "Schaeffler Buys Ewellix for €582m, to Boost Its Linear Range." https://drivesncontrols.com/schaeffler-buys-ewellix-for-e582m-to-boost-its-linear-range/ ↩