Planner result
State
RFQ-ready
Leg screen
268 N.m
Groups
5
Empty state: adjust any input to tailor the result, or use the defaults as a first RFQ baseline.
Tool-first hybrid guide
Estimate a humanoid actuator stack, see RFQ risk, then review the evidence, architecture tradeoffs, safety limits, and supplier questions behind the recommendation.
Route mode
Hybrid
Primary task
Stack + RFQ screen
Evidence date
2026-06-09
Defaults model a mid-size humanoid before detailed CAD and thermal data are available.
Live result
RFQ-ready: 268 N.m leg screen, 5 actuator groups
State
RFQ-ready
Leg screen
268 N.m
Groups
5
Empty state: adjust any input to tailor the result, or use the defaults as a first RFQ baseline.
This screen estimates actuator-family pressure, not final joint sizing. It intentionally separates body axes, leg peak events, upper-body payload work, and validation risk.
Upper-body screen: 73 N.m
Use this for shoulder/elbow/wrist routing only after payload and reach envelope are known.
Total moving mass: 62 kg
Mass includes payload because handling and recovery events alter leg and waist actuator load.
Contact: [email protected]
Include quantity, destination, and target sample date for a faster RFQ response.
Report summary
Use these numbers as public benchmark anchors. They establish a decision frame, but final actuator choice still needs supplier evidence for duty cycle, cooling, brakes, and lifecycle.
20-75 axes
Public examples span simple logistics arms through full body-and-hand stacks; actuator architecture must be planned by joint role, not by one motor family.
90-360 N.m
Unitree public data lists G1 knee torque at 90/120 N.m and H1/H2-class leg torque up to 360 N.m; continuous ratings still require supplier evidence.
3 g touch to 25 kg payload
Figure discloses fingertip sensing down to grams, while Apollo discloses a 55 lb payload; hand, arm, and logistics payload evidence are not interchangeable.
2025 + 2023
ISO 10218-2:2025 covers robot-cell integration, while ISO/PAS 5672:2023 addresses force and pressure measurements for human-robot contact.
Selected path: Quasi-direct drive. The chart shows screening tendencies, not a guaranteed supplier capability.
A robust RFQ turns public benchmarks into a supplier evidence package, then into pilot validation. Skip one layer and the result becomes procurement theater rather than engineering evidence.
| Step | Input | Output | Common failure |
|---|---|---|---|
| Map joint roles | Leg, waist, arm, wrist, and hand axis count | Actuator family split instead of one generic BOM line | Buying one torque class for every axis |
| Estimate peak envelope | Robot mass, payload, dynamic factor, lever-arm class | Screening torque for leg and upper-body actuator groups | Comparing only catalog stall or peak torque |
| Derate for continuous duty | Gait cycle, hold time, cooling path, enclosure temperature | Thermal evidence request for RFQ | Assuming peak torque density equals continuous capability |
| Select control topology | Backdrive need, impact tolerance, force-control bandwidth | QDD, geared, SEA, or custom branch | Choosing architecture before contact and impact tests |
| Close safety evidence | Contact scenario, brakes, stops, sensing, force measurement | Robot and cell-level validation plan | Treating actuator compliance as a safety certificate |
| Source | Signal used | Date / scope | Link |
|---|---|---|---|
| Unitree G1 product page | 23-43 degrees of freedom; single leg 6 DOF; knee torque 90 N.m / 120 N.m depending version; arm load about 2 kg / 3 kg. | Reviewed 2026-06-09 | Review |
| Unitree H1 / H1-2 product page | H1-2 lists 27 DOF, maximum arm joint torque 120 N.m, maximum leg joint torque 360 N.m, and 189 N.m/kg peak torque density. | Reviewed 2026-06-09 | Review |
| Unitree H2 Plus product page | Maximum arm torque 120 N.m, maximum leg torque 360 N.m, 7 kg rated arm payload, 15 kg peak arm payload, and 75 total body-and-hand DOF. | Reviewed 2026-06-09 | Review |
| Figure 03 product page | Figure lists an electric 5 ft 8 in humanoid with 61 kg weight, 20 kg payload, 5 hour runtime, and 1.2 m/s speed. | Reviewed 2026-06-09 | Review |
| Figure Helix 02 technical update | Figure states that whole-body policy outputs complete joint-level control and that fingertip tactile sensors detect forces as small as 3 grams. | Published 2026-01-27; reviewed 2026-06-09 | Review |
| Figure 02 BMW deployment retrospective | An 11-month BMW deployment reported 1,250+ runtime hours, 90,000+ parts loaded, and forearm failures tied to packaging, 3 DOF dexterity, and thermal constraints. | Published 2025-11-19; reviewed 2026-06-09 | Review |
| Apptronik Apollo product page | Apollo lists 5 ft 8 in height, 160 lb weight, 4 hour battery runtime, and 55 lb payload for warehouse and manufacturing use. | Reviewed 2026-06-09 | Review |
| Agility Robotics Digit launch and Toyota agreement | Digit was introduced with 4-DOF arms and up to 40 lb box handling; Toyota Motor Manufacturing Canada signed a 2026 RaaS agreement after a pilot. | 2019 launch and 2026-02-19 agreement; reviewed 2026-06-09 | Review |
| IFR World Robotics 2025 Service Robots executive summary | Professional service robot sales grew 9% in 2024; RaaS fleets grew 31% to more than 24,500 units, supporting serviceability and fleet uptime as actuator purchasing criteria. | Published 2025; reviewed 2026-06-09 | Review |
| ISO 10218-2:2025 | Robot applications and robot cells require integration, commissioning, operation, maintenance, and decommissioning safety controls. | Published 2025-02; reviewed 2026-06-09 | Review |
| ISO/PAS 5672:2023 | Specifies test methods for measuring and analyzing forces and pressures in physical human-robot contacts. | Published 2023-11; reviewed 2026-06-09 | Review |
| Bipedal Humanoid Hardware Design technology review | The review frames humanoid design as a holistic coupling between structure and actuator choice, with electric high-ratio reducers historically common for torque, speed, and size tradeoffs. | Published 2021; reviewed 2026-06-09 | Review |
| MIT Cheetah proprioceptive actuator paper | The paper links high torque density, high-bandwidth force control, and backdrivability to dynamic legged impact mitigation; it is legged-robot evidence, not a humanoid product guarantee. | Published 2017; reviewed 2026-06-09 | Review |
Unknowns: most public humanoid pages do not disclose winding temperature, continuous torque, drive current limits, gearbox lifecycle, lubrication, or exact control-loop bandwidth. These must be requested before final design lock.
Updated 2026-06-09. These examples show why humanoid actuator selection cannot be reduced to one torque number: logistics payload, hand sensing, runtime, and service model all change the actuator evidence package.
| Platform | Public data point | Actuator implication | Decision use |
|---|---|---|---|
| Unitree G1 / H-series | 23-75 disclosed DOF depending model and hand package; 90/120 N.m G1 knee references; H-class leg torque up to 360 N.m. | Good lower-body torque anchors, but continuous torque and exact cooling boundary remain supplier-confirmation items. | Use as a leg and body-axis benchmark, not as a universal actuator BOM. |
| Figure 03 / Helix 02 | Electric system; 61 kg robot; 20 kg payload; 5 hour runtime; tactile sensing as small as 3 g disclosed in Helix 02 update. | Shows that payload, fingertip force sensing, and whole-body control must be treated as one coupled stack. | Use to frame arm-hand sensing and runtime questions; actuator torque tables are not publicly disclosed. |
| Apptronik Apollo | 5 ft 8 in, 160 lb, 55 lb payload, 4 hour battery pack runtime, positioned for warehouse and manufacturing. | Payload and runtime claims make battery swap, joint thermal duty, and service access procurement issues. | Use for logistics payload comparison; detailed joint torque data is not public. |
| Agility Digit | 4-DOF arms in 2019 launch; up to 40 lb box handling; 2026 Toyota RaaS agreement after pilot. | A non-dexterous logistics arm can be the right actuator choice when workflow payload beats anthropomorphic hand fidelity. | Use as a counterexample to assuming every humanoid needs high-DOF hands. |
| Option | Best fit | Strengths | Limits |
|---|---|---|---|
| Quasi-direct-drive rotary joint | Hip, knee, ankle, shoulder programs needing torque transparency | Backdrive behavior, impact tolerance, force-control headroom | Large motor diameter, current demand, thermal path, brake strategy |
| Compact high-ratio geared actuator | Holding axes, compact elbows, wrists, and waist modules | High torque in smaller package and easier static hold | Reflected inertia, lower transparency, shock and backlash evidence |
| Series elastic actuator | Human interaction, compliant legs, collision-tolerant research | Embedded compliance and measurable spring deflection | Bandwidth, resonance, spring fatigue, larger package length |
| Linear actuator or tendon branch | Hands, knees with linkage geometry, or remote mass placement | Packaging freedom and force-path customization | Linkage nonlinearity, friction, cable stretch, maintenance burden |
| Dexterous hand micro-actuator stack | Fingers, thumb opposition, force-touch manipulation | High DOF density near contact tasks | Low torque scale, fragile geartrain, tactile calibration effort |
| Claim | Public data can support | Still needs supplier evidence | Status |
|---|---|---|---|
| Peak torque | Public comparison of rough leg and arm torque scale when the OEM publishes the number. | Continuous torque, RMS current, winding temperature, cooling boundary, bus voltage, and repeated-cycle derating. | Do not sign off from public pages alone. |
| Backdrivability / compliance | Architecture direction, such as QDD, SEA, high-ratio geared, or tendon/linear branch. | No-power backdrive torque, reflected inertia, friction, impact recovery, brake release logic, and control-loop bandwidth. | Treat as measurable, not a marketing adjective. |
| Human contact safety | Whether the actuator stack includes sensing, force limiting, brakes, or compliance features. | Application risk assessment, ISO 10218-2 cell integration, ISO/PAS 5672 contact force/pressure measurement, and residual-risk controls. | Actuator choice is evidence input, not safety certification. |
| Dexterous hand readiness | Hand DOF, tactile claims, task videos, or disclosed fingertip sensing thresholds. | Finger stall force, gear backlash, calibration drift, cable/tendon wear, thermal rise in forearm, and repair time per finger. | Public evidence often proves demos, not maintenance economics. |
| Fleet deployment cost | Public pilots, RaaS adoption, runtime hours, and broad service-robot market context. | Actuator replacement interval, field swap procedure, spare module pricing, warranty exclusions, and traceability of failed units. | Public evidence insufficient for TCO without supplier data. |
Pending confirmation: no reliable public source found during this 2026-06-09 review that discloses complete joint-by-joint continuous torque, winding temperature, gearbox life, brake fault-state, and repair interval for the benchmark humanoids above.
Upside: Fewer SKUs, simpler controller integration, easier inventory.
Downside: Oversized wrists or underspecified legs; thermal and mass penalties compound across 20+ axes.
Recommendation: Use one family only inside a joint group; keep legs, arms, wrists, and hands as separate evidence tracks.
Upside: Catalog-like modules reduce first-sample lead time and integration uncertainty.
Downside: Custom geometry may be required for mass placement, cable routing, brake location, or thermal path.
Recommendation: Run catalog and custom paths in parallel when planner output is Needs validation or Architecture risk.
Upside: Five-finger hands improve generality for tools, irregular objects, and bimanual tasks.
Downside: More actuators, tighter forearm packaging, more calibration drift, and higher repair burden.
Recommendation: Choose the simplest end effector that passes the target workflow; use high-DOF hands only when task variety justifies it.
Upside: High-ratio gearing can improve compact hold torque and reduce static current.
Downside: More reflected inertia and lower transparency can hurt contact-rich balance recovery.
Recommendation: Ask for measured impact recovery and no-power backdrive data before choosing high-ratio legs.
The tool is strongest during concept, RFQ, and supplier screening. It is not a replacement for detailed multibody dynamics, thermal modeling, or safety validation.
Probability: High | Impact: High
Ask for RMS current, winding temperature, cooling boundary, and repeated-cycle test data.
Probability: Medium | Impact: Medium
Split the actuator stack by joint role, duty cycle, and contact sensitivity.
Probability: Medium | Impact: High
Request no-power backdrive torque, reflected inertia, friction, and impact recovery tests.
Probability: Medium | Impact: High
Define hold torque, release logic, fault state, and manual recovery before sample build.
Probability: Medium | Impact: High
Run application-level risk assessment and contact force/pressure measurement where people can be contacted.
Stack: QDD knees/hips, compact wrist, optional dexterous hand
Gate: 90-120 N.m knee screening plus thermal walk-cycle evidence
Next: Start with G1-class public benchmark, then request continuous-duty data.
Stack: High-torque shoulder/elbow, geared wrist, brake-backed waist
Gate: Arm payload trace, brake fallback, fixture contact forces
Next: Treat H2 Plus arm payload as a public reference point, not a final spec.
Stack: Leg-dominant torque stack with impact and backdrive validation
Gate: Up to 360 N.m class leg screening plus shock and cooling tests
Next: Separate peak event, RMS gait, and hard-stop tests in the RFQ.
Stack: Arm actuator plus hand micro-actuator and tactile stack
Gate: Finger force, backlash, fingertip contact pressure, calibration drift
Next: Do not size the hand from body DOF alone; use object and contact cases.
Most modern humanoids combine rotary joint actuators for legs, waist, arms, and wrists with smaller hand actuators or tendon drives for fingers. The exact mix depends on torque density, backdrivability, brake strategy, cooling, and available package space.
No. The phrase actuators in humanoid robots belongs to the same decision cluster as humanoid actuator selection. The useful question is how the actuator stack changes by leg, waist, arm, wrist, and hand role.
Usually no. Legs, arms, wrists, and hands face different torque, speed, impact, and contact requirements. A single-family choice can simplify sourcing but often creates mass, thermal, or force-control compromises.
There is no universal range. Public references show smaller knee axes around 90-120 N.m and full-size leg-class peak torque up to about 360 N.m. Continuous duty, speed, and cooling must be validated separately.
This page is an actuator-stack planner for humanoid robots: it routes body axes into actuator groups, estimates RFQ risk, and compares architectures. A generic humanoid actuator page can cover definitions and product classes more broadly.
Choose it when torque transparency, impact tolerance, and force-control behavior are more important than the smallest possible package. It still needs thermal and brake validation.
It is often better for compact holding axes, wrists, elbows, and waist modules where static torque and package size dominate. The tradeoff is lower transparency and higher need for shock/backlash evidence.
Not always. Series elasticity helps with compliance, shock absorption, and force sensing, but it adds package length, resonance management, and spring fatigue validation.
No. Public specs are useful benchmarks, but they rarely disclose continuous torque, thermal boundary, lifecycle test setup, or exact safety case. Use them to frame RFQ questions, then require supplier evidence.
Digit shows that a humanoid form factor can prioritize payload handling with simpler 4-DOF arms instead of full anthropomorphic hands. That matters because the right actuator choice follows the workflow, not the human skeleton.
The missing public layer is usually continuous torque, winding temperature, drive current limits, reducer life, lubrication interval, brake fault behavior, repair time, and complete joint-by-joint duty-cycle data.
Fleet deployment raises the value of fast module swaps, fault traceability, spare pricing, and repair interval evidence. IFR 2025 service robot data shows RaaS fleet growth, so uptime can matter as much as peak torque.
Include robot mass, payload, joint axes, duty cycle, target torque/speed, package envelope, cooling assumptions, brake behavior, control interface, validation tests, quantity, destination, and timeline.
Treat safety as robot and application-level work. Actuator selection must support braking, stops, force limiting, and contact measurement, but standards and risk assessment apply to the integrated machine and task.
Send the computed inputs with your CAD envelope and intended motion cases. The minimum next path is a dual-track RFQ: one catalog-like joint route and one custom architecture route with explicit validation gaps.
Yes. The fastest path is to share joint-by-joint torque/speed targets, duty cycles, package constraints, and expected prototype quantity so feasibility feedback can be specific.
Send the tool summary plus joint CAD envelope, duty cycle, target quantities, and destination. We will route the request into actuator-family feasibility, sample path, and validation evidence.