BLOGS

Every humanoid spec sheet leads with peak torque. Physics does not care. A joint motor can hold its headline number for seconds before its copper windings cook; what it can sustain across a full working shift is a fraction of that — set almost entirely by how fast heat escapes a sealed limb. That fraction, and the cooling architecture that raises it, is quietly becoming the sharpest dividing line between demo robots and working ones.

Peak torque is a sprint. Continuous torque is a shift.

Peak torque answers a narrow question: can the actuator deliver one short, high-output event? It is limited by magnetics and drive current, and it is sustainable for seconds. Continuous (rated) torque is limited by heat: it is the highest output the motor can hold indefinitely without its winding temperature running away.

In air-cooled humanoid actuators, the continuous figure typically lands at just 25–30 percent of the peak figure — a gap engineers have started calling the thermal cliff, because it is where marketing meets physics. Published ratings for common quasi-direct-drive (QDD) actuators cluster around 2.5:1 to 3:1 peak-to-continuous ratios across knee, shoulder and wrist classes — and higher in some models — and even flagship platforms follow the pattern: one leading full-size humanoid publishes roughly 360 N·m of peak knee torque — but no continuous rating at all.

For sizing a real joint, neither number matters as much as a third one that rarely appears on the datasheet: the RMS torque of the actual motion cycle, computed over a full task and compared against the rated value with margin for hot ambient air inside the limb.

Where the heat comes from — and why a sealed limb traps it

Motor heat is mostly resistive: power lost as heat scales with the square of current (P = I²R). Double the load on a joint and the heat roughly quadruples. Worse, copper's resistance rises about 0.4 percent per degree Celsius, so a hot winding wastes more power than a cold one carrying the same current — a positive feedback loop that ends in thermal runaway if nothing intervenes. Winding insulation in typical joint motors starts to fail around 130–150 °C — the exact ceiling depends on the insulation class (class B is rated 130 °C, class F 155 °C) — and sustained overheating can permanently demagnetize the rare-earth magnets.

The cruel part for humanoids is that the worst case is not running — it is standing still. A motor holding a static pose produces torque at zero speed, so all electrical input becomes heat and none becomes mechanical work. A deep squat, a held box, a braced drilling posture: these are stall conditions, the most punishing state an electric motor knows.

A concrete worked case, from Firgelli's engineering guide to humanoid actuators: a knee actuator holding a 20 kg load in a bent pose needs about 80 N·m, but its derated continuous capacity inside a 40 °C sealed leg might be closer to 18 N·m. Meeting the pose means running at more than four times rated current, which multiplies heat generation roughly nineteen-fold — and puts the winding at its thermal limit in under two minutes. That arithmetic is why demonstration videos are short.

And unlike an industrial arm with a fan and open air around it, a humanoid actuator is sealed inside a limb — a totally enclosed, non-ventilated (TENV) package whose heat must conduct out through every interface in turn: copper windings, stator laminations, housing, limb structure, and finally the ambient air. Every interface adds thermal resistance. Heat that forced air would carry away in seconds takes minutes to conduct through solid metal.

Duty cycle compounds it. At a modest 1.4 steps per second, a walking humanoid puts on the order of 40,000 load cycles through its leg actuators over an 8-hour shift, each step carrying two to three times body weight. Heat that accumulates slightly faster than it escapes shows up an hour into the shift — not in a 30-second demo.

The cooling ladder: buying back continuous torque

Since the peak number is fixed by magnetics, the practical way to raise a joint's continuous rating is to remove heat faster. Each rung of the cooling ladder buys a bigger usable fraction of peak:

Ranges are indicative for comparable motor designs and trace to a single widely-cited engineering guide (Firgelli's), so treat each rung as an engineering rule of thumb rather than a measured constant.

The pattern is consistent across independent engineering analyses: moving from passive air to liquid can roughly triple the work a joint sustains without changing the motor, and liquid cooling removes heat several times faster — by some engineering estimates up to an order of magnitude — than air convection. The trade-offs are real — pumps, hoses, seals and heat exchangers all add mass, cost and potential failure points, and leak-safe dielectric fluids become a design requirement. Industry analyses of leading commercial humanoids describe liquid cooling loops serving the joint motors; the makers themselves have not confirmed the details, but the direction of travel across the industry is unmistakable.

There is also a cheaper rung most spec sheets ignore: don't generate the heat at all. Holding poses through brakes, non-backdrivable screws or series-elastic elements lets a joint carry static load with near-zero current, reserving the windings for motion.

2026: cooling moves inside the winding

The research frontier is now pushing heat removal past the housing and directly against the heat source.

The most striking example comes from a June 2026 study in *Applied Thermal Engineering*, in which researchers at Zhejiang University embedded porous phase-change hydrogel cooling channels directly into the stator slots of a humanoid joint motor — placing an evaporative heat sink beside the windings themselves. The hydrogel holds an evaporative water content of 86.7 percent and sustains compressive stresses of 4.0 MPa. In testing at 100 rpm under 6 N·m of load, the hydrogel-cooled motor's peak winding temperature dropped by 24.5 °C, and the time it could run before hitting its temperature ceiling extended by 43.1 percent, with temperature-induced torque fluctuation reduced as well.

Parallel work points the same direction: a 2026 study in *International Communications in Heat and Mass Transfer* modeled an annular liquid-cooling ring for high-performance actuators, keeping a 250 W peak-power motor below 85 °C where a passive simulation reached 380 °C, with a heat-transfer coefficient near 1,340 W/m²K at 0.6 L/min of coolant flow (simulation-led results, with a corrigendum issued since publication). And thermal awareness is entering control itself: a 2026 project from Huazhong University of Science and Technology folded a full 12-joint thermal model into reinforcement-learning training, so the robot moderates its own gait as joint temperatures climb.

The common thread: heat removal is becoming part of the joint architecture, not an accessory bolted on after the torque budget is spent.

What the thermal wall means for integrated joint modules

For integrated joint modules — motor, reducer, encoder, driver and structure packaged as one unit — thermal design stops being a motor problem and becomes a module problem:

• The thermal path is mechanical. Winding-to-stator-to-housing-to-mount interfaces decide the continuous rating before a single control loop is tuned. Potted windings and conductive housings shorten the path; air gaps anywhere along it throttle every joint downstream.
• Size by RMS, verify by soak. A module sized to peak specs will survive the demo and fail the shift. RMS torque of the real task cycle must sit below the continuous rating with margin for a 40–50 °C limb interior — and an hour of logged soak testing at target payload tells the truth no sprint test will.
• Sense temperature where it kills. Winding or housing temperature feedback, wired into the driver, lets a joint throttle or shed load gracefully before insulation damage or demagnetization — turning an invisible failure mode into a manageable derating curve.
• Hold with brakes, not current. Joints that lock mechanically under static load keep their thermal budget for motion — often the single highest-leverage feature for shift-length endurance.
• Ask where the continuous number was measured. A bench rating in free air is not the rating of the same module inside a sealed leg at 40 °C.

The five questions that beat a peak-torque table

For anyone evaluating humanoid actuators — or any robot meant to work rather than perform — the thermal questions reveal more than the torque table:

1. What is the continuous rating of the mounted joint, at a stated ambient temperature? Not the motor's bench rating in free air.
2. What is the RMS torque of my actual task cycle, and does it fit under the rated value with margin? Peak events only need to clear the peak rating briefly.
3. What happens in the static cases? Holding poses are stall conditions; ask how the design carries load without burning current.
4. What does the joint do at its thermal limit? Well-engineered joints throttle on temperature feedback with a published derating curve, rather than failing silently.
5. Do you have soak data, not sprint data? An hour of the target duty cycle with housing temperatures logged exposes the slow accumulation a demo never will.

Two robots with identical peak-torque tables can differ by 3–4× in the work they sustain, purely on cooling architecture. As humanoids move from pilots to shifts, the number deciding purchase orders is shifting with them: continuous torque per kilogram — not peak torque per kilogram.

Source: humanoid.guide, "The thermal wall: why continuous torque, not peak torque, decides what a humanoid can do" (humanoid-actuator-thermal-limits-continuous-torque, 2026)

Source: Firgelli, engineering guide to humanoid robot actuators (cooling ladder and stall-case arithmetic)

Source: Li, Q., Zhou, H., Mei, D., Wang, Y., "Highly efficient evaporative cooling of humanoid robot joint motors based on phase change hydrogels," Applied Thermal Engineering, Vol. 298 (June 2026)

Source: NE时代,《机器人运动会，跑出一门「液冷生意」》(2026-10)

Source: Intel Market Research, "Humanoid Robot Cooling System in 2026" (annular liquid-cooling ring study, 2026)

Source: evanino.com, "Humanoid Robotic Actuators: The Full Engineering Information" (TENV thermal path, liquid cooling comparison)

Source: e-eway,《人形机器人如何突破高功率密度伺服关节的散热与能效瓶颈?》