Three places the grease has to be

A strain-wave gear works by bending metal. An elliptical wave generator spins inside a thin flexible cup, the flexspline, and pushes its teeth into a rigid ring, the circular spline, at two points on the long axis of the ellipse. Those two mesh points travel round with the motor. So does everything that can go wrong with the lubrication.

A reliability study of harmonic drives in industrial robots, published through the PHM Society, names three zones that need a film of grease: the tooth mesh between flexspline and circular spline, the thin-section ball bearing on the wave generator, and the sliding contact between that bearing’s outer race and the inside of the flexspline. It singles out the last one. Wear there feeds particles into the grease, the particles spread through the gear, and efficiency drops. The same study ranks lubricant degradation among the higher-risk failure modes and points out the awkward part: the joint is sealed, so checking the grease means taking it apart.

Cycloidal and planetary reducers have their own versions of the problem – rolling pins, needle bearings and gear flanks instead of a flexing cup – but the logic is the same. A thin film separates metal from metal. When it thins out, gets contaminated or turns stiff, friction climbs and the joint pays for it in heat, lost torque and wear.

Rated for 0–40°C, running hotter

Here’s the mismatch. Harmonic Drive, the reference supplier for strain-wave gears, rates its standard SK-1A, SK-2 and SK-3 greases for 0°C to +40°C. Its catalog performance figures are quoted inside that same ambient band. The one grease it sells for wider conditions, 4B No.2, stretches to −10°C to +70°C.

Humanoid joints don’t stay in that band. In a 2026 bug report on NVIDIA’s GR00T whole-body controller, Unitree G1 owners logged ankle roll joints heating from 49°C to 80°C in 90 seconds and triggering the robot’s high-temperature warning at 90°C – one reached 96°C. With Unitree’s stock controller the same ankles sat at 30–45°C. That’s a reported actuator temperature, not a grease temperature, and the cause was a control policy loading the two ankle motors against each other. But it shows how quickly a small joint can leave the window its lubricant was specified for.

Heat costs grease life. Harmonic Drive’s CSD/SHD catalog plots the allowed number of input revolutions before a grease change against temperature, and the curves fall away sharply above 40°C. Load pushes the same way. The catalog scales grease life with the cube of the ratio between rated and average torque, LGT = LGTn × (Tr/Tav)³, so a joint running at twice its rated torque on average gets an eighth of the grease life.

Cold is the bigger efficiency hit

Hot grease wears out. Cold grease fights back. Viscosity climbs as temperature falls, and a strain-wave gear – which churns its lubricant twice per input revolution – feels that directly as drag.

The cleanest numbers come from space hardware, where engineers test lubricants at both extremes on purpose. At ESMATS 2015, Roberts and colleagues ran two identical HFUC-20-160 strain-wave gears at 4 N·m output and 50 rpm input, one lubricated with a PFPE oil and one with a multiply-alkylated cyclopentane (MAC) oil, cycling between +90°C and −25°C. Hot, both held around 80% or better. Cold, the PFPE gear dropped to about half. The MAC gear, whose oil has a higher pour point, started at around 20% and slid towards 10% by the end of the test.

Those are oils in a vacuum test rig, not greases in a walking robot. The industrial catalogs point the same way, though. Read off Harmonic Drive’s curves for its tiny CSF-3-100 gear on SK-2 grease, efficiency at 500 rpm input falls from roughly 60% at 20°C to about 45% at −10°C, and no-load running torque rises around five-fold. Nabtesco’s Molywhite RE No.00 cycloidal grease more than triples its starting torque in a low-temperature torque test between −30°C and −40°C.

Sealed for life – or just sealed

Industrial robot reducers are built to be regreased. Humanoid actuators mostly aren’t, at least not by anyone outside the factory. No major humanoid maker publishes a relubrication interval, which leaves two readings: the joint is designed to outlive its grease charge, or the grease change is folded into a module swap at service time. Either way the grease choice gets locked in at design.

That choice is a trade. Thin, low-viscosity greases cut drag and help cold starts but creep past seals – Harmonic Drive warns that 4B No.2 turns into a relatively thin fluid in operation and that some base-oil leakage at the seals can’t be ruled out. Stiffer greases stay put and protect under shock loads but cost efficiency. Nabtesco splits its RV range the same way, with a low-temperature grease (RV Grease LB00), an EP grease (Vigogrease RE 0), the older molybdenum-based Molywhite RE 00, and synthetic oils for high-speed duty. Klüber pitches its Klübersynth GE 44 series at harmonic reducers and GEM 4-150N at cycloidal gears.

Chinese reducer makers, now supplying much of the humanoid volume, are writing their own specifications. Shenzhen PICEA describes a target of −40°C to 120°C for robot harmonic-drive grease, built on synthetic PAO or trimethylolpropane ester with a 12-hydroxystearate lithium thickener at 5–20% of mass. Specialist SmartC sells a humanoid harmonic grease rated −40°C to +120°C and quotes a friction coefficient of 0.048 – the maker’s own figure, without published test conditions. Coating suppliers go at it from the other side: Ionbond markets DLC coatings for humanoid gears, pins and bearings as a way to reduce reliance on lubrication, though it publishes no numbers either.

What to ask an actuator supplier

• Which grease is in the reducer, and what temperature range is it rated for? If it’s 0–40°C, ask how hot the gear gets on your duty cycle.
• Efficiency and no-load torque at the lowest ambient you’ll run in, not just at 20–25°C.
• Is the grease charge meant to last the joint’s life? If not, what’s the interval, and is it a regrease or a module swap?
• Seal material and grease compatibility, especially for thin, long-life greases that weep at the seals.
• How the friction model in the drive handles temperature, if joint torque is estimated from motor current.

FAQ

Sources