Blog · 2026-07-05 · Servos
A hobby servo is a complete motion system in a plastic box — motor, gearbox, sensor and controller for a few hundred rupees. Here's how they work, how to read their specs, and how to stop killing them.
The hobby servo may be the most democratizing component in robotics: one three-wire connector and a single line of code produce controlled angular motion that would otherwise require a motor, gearbox, encoder, driver and a control loop you'd have to tune yourself. But that convenience hides real engineering variety — a ₹150 SG90 and a ₹2,500 serial-bus servo are profoundly different machines wearing the same silhouette. Understanding what's inside is what lets you pick correctly.
Four parts: a small DC motor; a gear train reducing its ~10,000+ RPM into slow, strong output; a potentiometer on the output shaft reporting the current angle; and a control board running a feedback loop — command angle in, pot reading compared, motor driven until the error closes. It's a complete closed-loop position controller (in fact, watching a P-controller work in our PID Tuning Visualizer is watching a servo's brain in slow motion). The standard interface is a 1–2 ms pulse repeated every 20 ms: 1.5 ms commands centre, 1.0 and 2.0 ms the extremes of a ~180° range. Three wires — ground, power, signal — and every hobby controller on Earth speaks it.
The distinction is the control board. Analog servos update their motor drive at the incoming 50 Hz pulse rate — simple, cheap, slightly soft in holding a position against disturbance. Digital servos run a microcontroller updating at 300+ Hz: they hold position stiffly, respond faster, and produce more torque near the target — at the cost of price and noticeably higher idle current (that faint constant buzz is the high-rate loop working). For arms and anything holding loads, digital earns its price; for a sensor-sweep bracket, analog is fine.
The gear train takes every shock the output shaft receives, so gear material is the reliability spec. Nylon/POM gears (SG90 class) are quiet and cheap but strip under shock — a falling arm back-driving the output is the classic killer. Metal gears (brass/steel/titanium mixes — the M in MG90S and MG996R) survive abuse that shreds plastic. Coreless and brushless motor variants at the premium end add speed and lifespan. The field rule is blunt: any joint that gravity can back-drive gets metal gears, full stop. The ₹80 saved on plastic is repaid the first time the arm drops.
Torque, in kg·cm, quoted at a test voltage — and it's a stall figure: the load at which the servo stops moving entirely. Size your working load to a fraction of it; that's the safety factor logic built into the Servo Torque Calculator, and the calculation method lives in the robot arm article. Speed, quoted as seconds per 60° (an MG996R's 0.17 s/60° means roughly half a second across full travel) — smaller is faster, and speed drops as load rises. Voltage range — most run 4.8–6 V; modern high-torque units accept 6.8–8.4 V and produce their headline numbers only at the top of the range. Both torque and speed scale meaningfully with voltage, which cuts both ways: a servo specced at 7.4 V delivers noticeably less on a 5 V rail. Dead band — the smallest command change the servo responds to; tighter dead band means finer positioning and more idle hunting.
| Class | Example | Torque (typ.) | Use it for |
|---|---|---|---|
| Micro, plastic | SG90 | ~1.8 kg·cm | Sensor sweeps, indicators, no gravity loads |
| Micro, metal | MG90S | ~2.2 kg·cm | Grippers, wrists, pan-tilt |
| Standard | MG996R | 9–11 kg·cm | Elbows, steering, small-arm shoulders |
| High-torque digital | DS3218 / DS3225 | 18–25 kg·cm | Desktop-arm shoulders, heavy grippers |
| Serial-bus "smart" | LX-16A, Dynamixel | 17–60+ kg·cm | Multi-DOF arms, hexapods, humanoids |
| Continuous rotation | FS90R | n/a (speed control) | Tiny drive wheels — a slow gearmotor in servo clothing |
Serial-bus servos deserve a note: they replace the PWM wire with a daisy-chained data bus, adding position/temperature/load feedback, configurable limits and IDs. Wiring a 6-DOF arm shrinks from a loom of PWM lines to one chain — past roughly four joints, they usually win on sanity alone.
The most common "broken servo" is a healthy servo on a starved supply. A single MG996R can spike 1–2.5 A under load; a six-servo arm moving briskly can demand 8–10 A — none of which can come from a microcontroller's 5 V pin. The symptoms are maddeningly indirect: controller brownouts and resets, servos twitching at startup, positions drifting under load. The fixes are standard practice: a dedicated 5–7.4 V supply (UBEC or buck converter) rated for the worst case, common ground between servo supply and controller, generous capacitance (470–1000 µF) across the servo rail, and a driver board like the PCA9685 to generate clean PWM for many servos — while its power still comes from the real supply, not USB.
Never force a powered servo's output by hand — you're fighting the loop and grinding the gears. Don't command positions the mechanics can't reach; a servo pinned against a hard stop is at continuous stall, cooking itself (buzzing at rest is the audible warning). Use the mounting flanges and rubber grommets — the case isn't a structural member. And centre servos in software before assembling horns, so the mechanical range lands where the electrical range is.
Servos know where they are (closed loop) but offer ~180° and modest resolution; steppers offer fine open-loop resolution over unlimited rotation but lose position if overloaded. Arms → servos; plotters and sliders → steppers.
Some servos are sold as 270°; serial-bus servos often do multi-turn. Gearing a standard servo up externally multiplies angle at the cost of torque — the same trade as ever.
In rough order of likelihood: noisy or shared power supply, missing common ground, unstable PWM from a software timer (use hardware PWM or a driver board), or a worn potentiometer in an old unit.
Class chosen, specs decoded — the remaining step is arithmetic: confirm the torque number with the Servo Torque Calculator before you order. Two minutes of division is the difference between a joint that lifts and a joint that hums.
Servo leads are almost standardized: three pins at 2.54 mm — ground, positive, signal — with two dominant color schemes. Futaba-style: black/red/white. JR/Hitec-style: brown/red/orange. Signal is always the lightest wire, ground the darkest, power in the middle — a layout chosen so a reversed connector puts signal on ground harmlessly rather than power on signal fatally. The real trap is voltage, not polarity: plugging a 6 V-rated servo into a 2S LiPo rail (7.4–8.4 V) works impressively for a while, then releases the smoke. Check the servo's rated range before it meets any new rail, and label rails in multi-voltage robots. If a servo lead must run long (>30 cm) near motor wiring, twist it or add a small capacitor at the servo — induced noise on the signal line is a classic source of one-servo-in-six jitter.
A servo buried in an assembled arm is annoying to replace, so audition it first: five minutes with a servo tester (₹150) or a ten-line sketch. Sweep the full range and listen — a healthy servo sounds smooth and even; grinding or a catch at one spot is gear damage. Command the centre repeatedly from both directions and watch for it landing in the same place — offset returns reveal gear backlash or a worn pot. Load it by hand (unpowered!) and feel the gear lash. Finally, hold a light load at horizontal for a minute and touch the case: warm is normal, hot means this unit shouldn't get a structural job. Batch-buy hobby servos vary enough that testing six and assigning the best two to the shoulder joints is a genuinely effective quality strategy — free performance from ten minutes of triage.
Spend on the joints that gravity loads and the joints that are hard to reach after assembly; economize on sweeps, pans and indicators. In practice that means metal-gear everywhere structural, digital at the shoulder, and the cheap plastic classics reserved for jobs where their failure costs a smile rather than a rebuild.
And a last word on counterfeits, because the servo market is thick with them: bargain "MG996R" units with plastic gears inside metal-look cases are common enough that the bench test above doubles as counterfeit detection. Weight is the quick tell — genuine metal-gear standards run around 55 g; a suspiciously light one is telling you what's inside. Buy load-bearing servos from suppliers who stand behind returns, and let the marketplace lottery supply only the joints that don't matter.