A 20 kg-cm servo: measured under what conditions?
Servos often account for more than half the bill of materials of a desktop robot, and their headline number is the least comparable figure in the whole list. Align the conditions before comparing — and the ranking often changes.
- Typical model
- Feetech STS series (used by SO-100, SO-101, LeKiwi)
- Price band
- Under about USD 20 per unit
- Communication
- Bus (half-duplex serial), several units daisy-chained
- Feedback
- Position feedback — you can read the current angle back
- Gearing
- Mostly metal gears; differs by model, check the datasheet for the specific one
- The spec, and what it measures
- The datasheet says 20 kg-cm, but that number is missing at least four preconditions: at what voltage, stall or rated torque, sustained for how long, and measured at what arm length and by what method. Vendors differ, and two servos of the same nominal rating re-measured on the same terms can swap order. Without the preconditions a torque figure is an order-of-magnitude hint, not a ranking.
- Price tiers
- The low band (teens to low twenties of dollars per unit) is the current default across the DIY ecosystem: bus communication plus position feedback, a combination that cost much more a few years ago. What you pay for it is durability and consistency — units within one batch vary, so a multi-joint build needs calibration margin and spares. Moving up a band mostly buys durability and consistency rather than double the torque.
- How to pick
- Four steps, in order: (1) open the datasheets side by side and align voltage, stall versus rated, and duration; (2) look at gear material and backlash, which datasheets often omit but which decide life under repeated impact; (3) confirm the protocol and host-side support — ecosystem lock-in on bus servos affects your schedule more than torque does; (4) price the whole joint count, because servos dominate the bill and a few dollars per unit multiplies.
- Known pitfalls
- (1) Ranking by the datasheet number, the most expensive habit here, especially on a multi-joint machine. (2) Ignoring the voltage convention: the same servo delivers very different torque at different voltages, so cross-voltage comparison is no comparison at all. (3) No spares: consistency at this price means occasional outliers, and testing units as a set before assembly is cheaper than reworking afterwards.
- What we checked
- We checked that this family is treated as the default choice across several open hardware projects, that bus communication and position feedback are genuinely present, and the price band. Not checked: we have not run a torque bench test on specific models under a common protocol — so what is offered here is a method for aligning conditions, not a measured ranking of two particular servos.
This site's call
Do not rank by the datasheet number; rank by the conditions the datasheet specifies. This is the cheapest single discipline in the whole parts list, because servos are such a large share of the cost that the error multiplies with joint count. Twenty extra minutes aligning conditions usually saves the price of a spare set.
Align four things before comparing
- Voltage: the same servo differs substantially across voltages; comparing across them is meaningless.
- Stall versus rated: stall is the force when it cannot turn, rated is what it can sustain. The second is closer to your actual need.
- Duration: holding peak torque for seconds is a different specification from continuous output.
- Gear material and backlash: the durability figures that are not numbers. Metal and plastic gears behave very differently under repeated impact, and datasheets rarely say so.
Once these four are aligned, the candidate list usually reorders relative to the datasheets. The money saved typically covers a spare set.
Sources
- Side-by-side comparison of torque test conditions in mainstream servo datasheets (voltage, stall vs rated, duration)
- Published bills of materials from open hardware projects (confirming where this series is actually used)
Objects in this entry
Last checked 2026-09-28