robotic-actuator

Robotic actuator

Component Hard import robotics
TRL Mars
Energy intensity
Required by
0
Requires
3

Electromechanical actuator providing joint torque + position control. Five mature and cutting-edge variants span the design space: BLDC + harmonic drive (industrial standard, Optimus), BLDC + cycloidal (Unitree, robust to overload), series elastic actuator (SEA, Pratt + Williamson 1995, compliant and proprioceptive), quasi-direct-drive (MIT Cheetah heritage, low gear ratio + high-torque motor), and soft-actuator (TCP twisted-coiled polymer + twistron CNT-yarn, biological-muscle-class power density). Mars chooses by dust tolerance + cold-soak + radiation; QDD wins.

Last reviewed: 2026-06-09

Governing equations

Geared actuator output torque = gear ratio × motor torque × efficiency. Harmonic drive: N = 100, η = 0.8; cycloidal: N = 50, η = 0.9. [1]

Output speed inversely with gear ratio. Same motor at higher N → higher torque, lower speed. Trade against backdrivability. [1]

Twisted-coiled polymer actuator force ∝ temperature change × polymer stiffness. Joule-heating of nylon fishing line creates contractile force comparable to skeletal muscle (Haines 2014 Science). [2]

Twistron carbon-nanotube-yarn torsional actuator specific power. Foroughi 2011: 60× human muscle; Mu 2019: 250 W/kg sustained — competitive with BLDC motors at tiny scale. [3]

End-to-end electrical-to-mechanical efficiency. BLDC controller: 95-98 %; motor: 85-95 %; harmonic drive: 80 %. Total: 65-80 %. [4]

Key constants & quantities

Symbol Value Units Conditions Description
τ_optimus-knee 50 ±15 Nm Nm peak (Optimus knee joint, Gen 2 estimate) Estimated peak torque at humanoid knee joint based on standing + step-up tasks. Mars-g reduces peak demand to ~ 20 Nm.[5]
P_specific,BLDC 500–1500 W / kg (BLDC motor specific power) High-power BLDC motors used in modern robots — competitive with hydraulics at this scale.[4]
τ_per_unit-mass,harmonic 80 ±20 Nm/kg Nm / kg (harmonic drive output) Harmonic drive torque-per-mass at industrial scale. Lower than cycloidal but higher precision.[1]
τ_per_unit-mass,cycloidal 150 Nm / kg (cycloidal drive output) Cycloidal drive torque-per-mass. ~ 2× harmonic; 5x overload tolerance vs harmonic drive.[1]
τ_per_unit-mass,QDD 50 Nm / kg (quasi-direct-drive) QDD architecture uses larger pancake motor + low gear ratio (3–8:1). Lower torque density but excellent backdrivability + proprioception.[4]
P_specific,muscle 200 W / kg (skeletal muscle peak) Human skeletal muscle peak specific power. BLDC motors at high-power exceed by 5-10×; biological actuator efficiency advantage is integrated over many parameters.[4]
cycles_to_failure,harmonic 100,000,000 cycles (harmonic drive nominal) Harmonic drive cycle life under nominal load. Order of 100 million cycles before lubricant + wear-element replacement.[1]
τ_TCP,Haines 4 MPa (twisted-coiled polymer tensile stress) TCP actuator tensile stress in nylon-fishing-line variant (Haines 2014 Science). Stress comparable to natural muscle; strain 49 % linear contraction.[2]

Operating envelope

ParameterRangeUnitsSource
Operating temperature (BLDC + electronics) -40 – 125 °C (industrial-grade) [4]
Operating temperature (Mars-rated) -90 – 60 °C (after thermal conditioning) [6]
Gear ratio (Mars robotic joint) 3 – 100 :1 [1]
Peak torque overload tolerance 1.5 – 5 × nominal [1]
Bandwidth (control loop) 100 – 1000 Hz [4]

Mass balance

Basis: 1 humanoid robot total actuator inventory (~ 28 joints)

Inputs

BLDC motor + controller per joint 1.5 kg each (Optimus-class) [5]
Gear set (harmonic + cycloidal mix) per joint 0.5 kg each [1]
Joint structural mass 0.5 kg each [5]
  • BLDC motor + controller per joint: ~ 42 kg of motor + electronics distributed across 28 joints; ~ 60 % of robot mass.
  • Gear set (harmonic + cycloidal mix) per joint: ~ 14 kg of gear sets.
  • Joint structural mass: Bearings, frame, sealing.

Outputs

Total joint torque inventory 1,400 Nm cumulative across 28 joints [5]
Operational lifetime 50,000 h between major actuator overhauls [5]
TRL · Earth
9/ 9
TRL · Mars
5/ 9
BLDC + harmonic / cycloidal drives: TRL 9 — used industrially since 1980s. Quasi-direct-drive (MIT Cheetah heritage): TRL 8 — fielded in Spot, Cassie, Digit. Twisted-coiled polymer (Haines 2014): TRL 4–5 — university research scale. Twistron CNT-yarn (Foroughi 2011, Mu 2019): TRL 3–4 — lab demonstrations. On Mars: TRL 4–5 — dust + cold-soak engineering required even for industrial-mature variants; Mars-flight unit yet.[4]
Energy budget
0 kWhe / component (energy use is at robot level) [4]

Actuator efficiency 65–80 % converts to robot kWh/h on the parent robot. End-to-end: ~ 30 % of electrical input becomes mechanical work; rest is heat.

Variants & trade-offs

BLDC + harmonic drive (industrial / Optimus / Apollo)

[1]

Brushless DC motor + harmonic-drive (strain-wave) speed reducer. High gear ratio (50–160:1), zero backlash, smooth output. Industry standard for industrial robots since 1957.

Gear ratio
50–160 :1
Efficiency
75–85 %
Backlash
0–0.1 arcmin
Stack lifetime
30000–60000 h to first overhaul
Materials: Steel flex-spline + circular spline · Wave generator (bearing on elliptical cam) · Vacuum-stable lubricant · Industrial-grade BLDC motor
  • Highest precision (zero-backlash)
  • Mature industrial supply chain
  • High torque density (compact for given output)
  • Predictable + well-modeled wear curves
  • Overload sensitivity (flex-spline can crack at > 2× rated torque)
  • Lubricant viscosity sensitive to Mars cold
  • Higher cost per Nm than cycloidal

BLDC + cycloidal drive (Unitree / DEEP Robotics)

[1]

Brushless DC motor + cycloidal (cyclo) gear. Multiple contact points distribute load → high overload tolerance. Used heavily in Chinese humanoid + quadrupedal robots.

Gear ratio
29–100 :1
Efficiency
85–92 %
Overload tolerance
3–5 × rated torque
Stack lifetime
40000–80000 h
Materials: Cycloidal eccentric cam + output rollers · Standard industrial BLDC motor · Synthetic oil lubricant
  • Overload-tolerant (3-5× rated)
  • Higher efficiency than harmonic
  • Lower cost (used industrially since 1930s)
  • Robust to shock loading
  • Small backlash (vs harmonic's zero)
  • Slightly larger physical footprint per Nm
  • Vibration at higher speeds

Quasi-direct-drive (MIT Cheetah / Spot / Cassie / Digit)

[4]

Large pancake-style BLDC motor + low gear ratio (3-8:1). Trades torque density for backdrivability + proprioception. Pratt + Williamson conceptual lineage; MIT Cheetah hardware breakthrough.

Gear ratio
3–8 :1
Backdrivability
0–0 High (touch-sensitive)
Joint bandwidth
300–1000 Hz
Stack lifetime
40000–80000 h
Materials: Pancake BLDC motor (large diameter, thin axial) · Low-ratio epicyclic gear · High-current motor controller · Standard ball bearings
  • Backdrivable — robot responds to applied torque (safety + manipulation)
  • High bandwidth — fast dynamic response
  • Lower wear-element count → longer lifetime
  • Best proprioceptive feedback (current sensing = applied torque)
  • Lower torque density per kg actuator
  • Higher motor cost per Nm output
  • Requires high-current motor controller

Twisted-coiled polymer (TCP, Haines 2014)

[2]

Twisted-spun nylon or polyester fiber wrapped tightly around itself. Joule-heating via embedded conductor contracts the fiber (~ 49 % linear strain). Power density ~ 100x natural muscle.

Tensile stress
1–10 MPa
Linear contraction
20–49 %
Cycle frequency
0.1–5 Hz
Stack lifetime
1000–10000 cycles before degradation
Materials: Nylon-6,6 or polyester fishing-line analog · Embedded silver or carbon-fiber conductor · Twist-and-coil manufacturing
  • Extremely low cost (~ $0.01/cm of actuator)
  • High specific energy (~ 100x muscle)
  • Soft + safe (no rigid components)
  • Massive parallelization possible
  • Low bandwidth (thermal cycling limited)
  • Joule heating efficiency low (energy as heat, not work)
  • Cycle fatigue under repeated loading
  • Single-use orientation (twist direction matters)

When preferred: Soft robotics, prosthetics, distributed actuators, soft EVA suit applications.

Twistron CNT-yarn (Foroughi 2011, Mu 2019)

[3]

Twisted carbon-nanotube yarn that generates rotational + tensile force when stretched, heated, or chemically activated. Thread-based motors with 250 W/kg specific power; competitive with BLDC at tiny scales.

Specific power
100–250 W/kg
Strain
5–30 %
Operating mode
0–0 Tensile / torsional / thermal
Stack lifetime
500–5000 cycles
Materials: Carbon nanotube yarn (multi-wall) · Electrolyte coating (electrochemical variant) · Stretch + twist apparatus
  • Highest specific power of any soft actuator
  • Multiple actuation modes (tensile, torsional, thermal)
  • Can be miniaturized to thread-scale
  • Continuous + reversible
  • TRL 3-4 — university research only
  • CNT supply chain immature for industrial scale
  • Limited cycle life vs traditional actuators
  • Material costs higher than polymer alternatives

When preferred: Future high-performance soft robotic skin, prosthetics, fine-motor microfluidics; not near-term Mars-base humanoid joint.

Failure modes

Mode Cause Detection Mitigation
Lubricant cold-soak viscosity rise[7] Mars night T (−90 °C) raises gear lubricant viscosity 1000x; cold-start torque spikes; motor stall risk. Pre-startup torque test; current trace at cold-start. Mars-cold-rated synthetic lubricants (PFPE, perfluoropolyether); pre-EVA actuator warm-up cycle; insulated actuator housing.
Dust ingress to gear contact surfaces[7] Perchlorate-rich Mars regolith enters past seals; abrasion + chemical attack on gear teeth + bearing races. Vibration signature change; current at constant load rises; gear-tooth wear via accelerometer harmonics. Multi-stage labyrinth seals; positive-pressure actuator housings (purge with clean gas during EVA); field-replaceable actuator modules.
Harmonic drive flex-spline crack[1] Overload event (> 2× rated torque) cracks the flex-spline; joint becomes inoperable. Sudden joint failure; gear-mesh signature loss. Conservative torque limits (1.5× design); cycloidal drive variant for high-shock applications; field-replaceable flex-spline modules.
Motor winding short circuit[4] Insulation breakdown under thermal cycling or moisture ingress; motor fails open or short. Phase current imbalance; controller fault. Sealed motor enclosure; thermal-cycle-rated insulation; redundant 3-phase windings; current monitoring + auto-disable.
Motor controller SEU[6] Mars-surface GCR + SPE causes single-event upsets in motor controller silicon; current spike or torque transient. Controller self-test; redundant-controller mismatch. Mars-radiation-rated MOSFETs + gate drivers; ECC firmware memory; watchdog reset; in-habitat operation reduces dose exposure.
Encoder failure[4] Optical encoder degraded by dust on optical disk; magnetic encoder degraded by stray field. Position-feedback discontinuity; motor control oscillation. Sealed encoder housing; redundant encoders (one optical + one magnetic); accept some accuracy loss in degraded modes.
Polymer fatigue (TCP / twistron variants)[2] Repeated thermal-strain cycles fatigue polymer or CNT yarn; output force degrades. Force-output trend; cycle-counter limit. Modular replaceable polymer / yarn elements; conservative cycle limits; sacrificial design.

Mars adjustments

Cold-soak operation viability[7]

Impact: Mars night T (−90 °C) exceeds typical electronic + lubricant operating range. Cold-start torque spikes; motor cogging worse; controller silicon at edge of operational envelope.

Mitigation: Mars-cold-rated lubricants (PFPE class); pre-conditioning heater cycle before operation; sealed + heated motor housings; LFP-chemistry batteries for motor power.

Dust ingress at every seal interface[7]

Impact: Perchlorate-rich Mars dust at every joint interface. Apollo LRV wheel bearings seized within hours of EVA exposure; Mars adds chemistry to abrasion.

Mitigation: Multi-stage labyrinth seals; positive-pressure actuator housings with clean-gas purge during EVA; field-replaceable actuator modules.

Radiation tolerance of controller silicon[6]

Impact: Motor controller silicon experiences cumulative TID dose + SEU rate ~ 10x Earth LEO. Standard COTS parts degrade faster than design life.

Mitigation: Mars-radiation-rated controllers (RAD-tolerant MOSFETs); TMR critical path; periodic safe-mode reset.

Lower gravity reduces actuator burden[4]

Impact: 0.38 g reduces joint torque by 62 % vs Earth. Same actuator effectively does 2.6× the work, enabling longer duty cycles, higher payload, or smaller / lighter actuator design.

Mitigation: Real benefit — Mars-designed humanoids can be lighter or more capable per Nm of actuator. Mass margin recovered for radiation hardening + dust seals.

In-situ manufacturing of replacement parts[6]

Impact: Mass-launched actuator inventory is finite. Long-duration colony requires on-site manufacturing of replacement gears, bearings, motor windings.

Mitigation: 3D-printed Inconel + polymer parts via Mars-base additive manufacturing; modular actuator architecture; commonality across robot fleet for spare-part economies.

Alternatives & substitutes

Hydraulic actuator (Atlas hydraulic / industrial)[4]

  • Highest peak force per mass
  • Compact for high-force tasks
  • Mature industrial heritage
  • Heavy + complex hydraulic infrastructure (pumps, valves, fluid)
  • Fluid leakage in vacuum / Mars-cold
  • Boston Dynamics abandoned hydraulic Atlas in 2024 for electric
  • Higher energy losses than electric actuators

When preferred: High-force industrial tasks; not modern humanoid.

Pneumatic / McKibben artificial muscle[4]

  • Soft compliance, safe for human interaction
  • High peak force per mass
  • Simple architecture (pressure → force)
  • Requires compressor + plumbing (mass overhead)
  • Compressed gas storage on Mars complex
  • Slow bandwidth vs electric

When preferred: Soft prosthetics, soft EVA suit augmentation; not primary humanoid joint.

Requires

References

  1. Schafer, I., Bourlier, P., Hantschack, F., et al. (Harmonic Drive Systems Inc.) (2019). Engineering Data for Harmonic Drive Components. Harmonic Drive LLC / Sumitomo Heavy Industries. HD-D2019. — Industrial reference for harmonic drive + cycloidal drive engineering: torque ratings, efficiency, cycle life, lubrication.
  2. Haines, C. S., Lima, M. D., Li, N., Spinks, G. M., Foroughi, J., Madden, J. D. W., et al. (2014). Artificial Muscles from Fishing Line and Sewing Thread. Science, 343(6173), 868-872. doi:10.1126/science.1246906 — Foundational paper on twisted-coiled polymer (TCP) actuators. Joule-heated nylon-fishing-line + sewing-thread fibers as artificial muscle; ~ 100× human muscle specific power.
  3. Foroughi, J., Spinks, G. M., Wallace, G. G., Oh, J., Kozlov, M. E., Fang, S., et al. (2011). Torsional Carbon Nanotube Artificial Muscles. Science, 334(6055), 494-497. doi:10.1126/science.1211220 — Twistron CNT-yarn rotational actuator. Mu 2019 (Science 365:150-155) follow-up: 250 W/kg specific power, electrochemical + thermal activation modes.
  4. Pratt, G. A., & Williamson, M. M. (1995). Series Elastic Actuators. IEEE/RSJ International Conference on Intelligent Robots and Systems. doi:10.1109/IROS.1995.525827 — Foundational paper on Series Elastic Actuators (SEA). Lineage to MIT Cheetah, Cassie, Digit, modern quasi-direct-drive actuator architectures.
  5. Tesla Robotics + Figure AI + Apptronik + Agility Robotics (2024). Humanoid Robotics 2024: Optimus Gen 2 / Figure 02 / Apollo / Digit — Public Specifications and Industrial Deployments. Tesla / Figure / Apptronik / Agility public statements. — Tesla Optimus Gen 2 (Dec 2023 reveal), Figure 02 (BMW Spartanburg deployment Aug 2024), Apptronik Apollo (Mercedes-Benz pilot 2024), Agility Digit (Amazon warehouses 2024). Cross-referenced via public IAC + earnings call statements + industrial pilot data.
  6. Drake, B. G. (Ed.) (2009). Human Exploration of Mars: Design Reference Architecture 5.0. NASA Johnson Space Center, NASA SP-2009-566. NASA/SP-2009-566. — NASA Mars Design Reference Architecture 5.0; mission architecture, MAV reference designs, ISRU mass budgets.
  7. Davila, A. F., Willson, D., Coates, J. D., & McKay, C. P. (2013). Perchlorate on Mars: a chemical hazard and a resource for humans. International Journal of Astrobiology, 12(4), 321-325. doi:10.1017/S1473550413000164 — Biological reduction of perchlorate as pre-treatment for ISRU water.