[Beginner] — Motors, drivers, gearing, and the current draw that makes power design suddenly matter.
The choice
Actuator selection determines what your robot can do more than any software decision. An undersized motor cannot be fixed in firmware.
| Type | Control | Feedback | Typical use |
|---|---|---|---|
| Brushed DC | voltage/PWM → speed | add your own encoder | cheap drive wheels, hobby |
| Brushless (BLDC) | commutated phases | Hall sensors or sensorless | drones, e-bikes, modern arms |
| Stepper | step pulses → discrete position | usually none (open loop) | 3D printers, CNC, positioning |
| RC servo | PWM pulse width → angle | internal, hidden | hobby arms, small joints |
| Smart servo (Dynamixel) | packet protocol | position, speed, load, temp | education, research arms |
Brushed DC
The simplest thing that moves.
Voltage roughly sets speed; current roughly sets torque. The two relations worth carrying:
The second is why a motor draws stall current when held still: means no back-EMF, so the only thing limiting current is the winding resistance, which is small. Stall current is often 10–20× the running current, and it’s what melts drivers and browns out your logic supply.
Controlled by PWM through an H-bridge — four switches giving forward, reverse, brake and coast.
What bites:
- Brushes wear out. Finite life, and they throw electrical noise
- Inrush and stall current. Spec the driver for stall, not for nominal. Current-limit in hardware if you can
- Back-EMF spikes. A motor is an inductor; switching it off produces a voltage spike. Flyback diodes, and keep motor wiring away from signal wiring
- A motor is a noise source. Separate power rails, generous decoupling, and a common ground star point. → Electricity
The single commonest beginner failure is powering motors from the microcontroller’s supply. The MCU browns out on the first stall, resets, and the fault looks like a software bug. Separate supplies, common ground.
Brushless
Better in almost every way, and more complicated to drive.
No brushes — the commutation is done electronically, by an ESC energising three phases in sequence. Higher efficiency, higher power density, much longer life.
Sensored uses Hall effect sensors for rotor position — smooth from zero speed, needed for anything that must produce torque at standstill (a robot joint, a wheel on a hill).
Sensorless infers position from back-EMF, which doesn’t exist at zero speed. Fine for propellers, bad for joints — a sensorless BLDC has no reliable low-speed torque, which is exactly what a robot arm needs.
Field-oriented control (FOC) is the modern way to drive them: transform the three phase currents into torque-producing and flux-producing components and control those directly. Smooth torque, high efficiency, precise control at all speeds. Available now on cheap hardware (ODrive, SimpleFOC, VESC), which is why hobby-level robot joints got dramatically better in the last decade.
Steppers
Move in discrete steps — typically 1.8° (200 per revolution), and microstepping subdivides further.
The appeal: position control with no encoder. Count the steps you sent and you know where you are — that’s why 3D printers and CNC machines use them, and why they’re cheap to integrate.
The catch, and it’s a big one: that’s open loop. Push too hard, accelerate too fast, or hit an obstacle, and the motor skips steps silently. The controller believes it’s at position 100 while the machine is at 94, and nothing reports the discrepancy. → Open loop vs closed loop
Also:
- Torque falls off sharply with speed. Steppers are strong slow and weak fast
- They draw full current even at standstill, so they run hot and waste power
- Microstepping gains smoothness, not accuracy. 1/16 microstepping does not give you 16× the positional accuracy — friction and detent torque dominate well before that
Closed-loop steppers (a stepper plus an encoder plus a servo controller) fix the skipping problem and are worth the small premium on anything that matters.
Servos
RC servos — a DC motor, gearbox, potentiometer and controller in one box. Command an angle with a PWM pulse width (1–2 ms), and the internal loop holds it.
Trivially easy, and you get no feedback out. You command 90° and have no idea whether it got there, or whether it’s straining. Fine for hobby, unacceptable where it matters.
Smart servos (Dynamixel, LX-16A, Feetech) talk a packet protocol over a serial bus and report position, speed, load, voltage and temperature. Daisy-chainable, individually addressable. This is the right choice for a first real arm — the difference between “I told it to go there” and “it’s there, drawing 300 mA, and it’s warm” is enormous when debugging.
Gearing
Motors are fast and weak; robots need slow and strong. Almost every joint has a gearbox, and the reduction is usually 50:1 to 300:1.
| Type | Backlash | Efficiency | Note |
|---|---|---|---|
| Spur | moderate | high | cheap, noisy |
| Planetary | low–moderate | high | compact, high torque, common |
| Harmonic drive | near zero | moderate | precision arms; expensive |
| Cycloidal | low | high | high ratio, high shock tolerance |
| Worm | low | low (~50%) | self-locking — holds without power |
| Belt/cable | low | high | moves the motor mass off the joint |
Backlash is the one that ruins control. Free play between gear teeth means the output doesn’t move when you reverse direction — until it does, suddenly. It’s a hard nonlinearity, it cannot be tuned away, and it causes limit cycles in a position loop. → Nonlinear Control
Which is why precision arms use harmonic drives despite the cost, and why a cheap gearbox caps your achievable accuracy no matter how good your controller is.
Two more things worth knowing:
Reflected inertia scales as . A high reduction makes the load feel enormously heavier to the motor, which is good for holding a load and bad for backdrivability.
Backdrivability matters for safety. A worm gear self-locks — it holds position with no power, which is great for a lift and terrible for a collaborative arm, because a human cannot push it out of the way. Non-backdrivable joints and humans are a bad combination. → Safety
Drivers
The layer between logic and power, and the thing that burns out.
H-bridge for brushed DC (L298N is the classic teaching part and is genuinely poor — old bipolar design, ~2 V dropped, gets hot; DRV8833/TB6612 are strictly better for small motors). ESC for brushless. Stepper drivers (A4988, TMC-series — the Trinamic parts are dramatically quieter).
What to check before buying:
- Continuous and peak current rating. The headline number is usually peak
- Voltage range, with margin for back-EMF spikes
- Current sensing, if you want torque control or collision detection
- Thermal protection and overcurrent limiting — a driver that shuts down beats one that fails short
- Logic level — 3.3 V vs 5 V, and whether it’s tolerant
Power
The part that’s easy to underestimate and hard to retrofit.
Budget for stall, not for nominal. Six joints stalling simultaneously is a real scenario — it’s what happens when the arm hits something.
Battery chemistry:
| Energy density | Note | |
|---|---|---|
| LiPo | high | high discharge, needs a balance charger, fire risk if abused |
| Li-ion (18650) | high | safer, lower discharge rate, cheap in packs |
| LiFePO4 | moderate | very safe, long life, heavier |
| NiMH / SLA | low | cheap, heavy, mostly obsolete |
Never charge or discharge lithium cells outside their spec, and use a protection/BMS board. This is the one place in robotics where a mistake starts a fire.
Separate the rails. Motor power and logic power on different regulators, joined at a single ground point. A motor stalling should not reset your controller.
And check your wire gauge. 20 A through a thin wire is a heater, and voltage drop over a long run to a distant joint is real.
From command to motion
Tying it together — the chain from “move there” to actual movement:
trajectory → joint setpoints → [POSITION LOOP] → velocity cmd
↓
[VELOCITY LOOP] → current cmd
↓
[CURRENT LOOP] → PWM → motor
Three nested loops, each faster than the one outside it — typically 100 Hz / 1 kHz / 10–20 kHz. That’s cascade control, and it’s how essentially every robot joint is driven. The inner loops absorb disturbances and linearise the actuator before the outer loop ever sees them. → Robot Control
The physical limits your planner must respect:
- Maximum torque — sets acceleration
- Maximum speed — sets traverse time
- Thermal limits — you can exceed rated torque briefly, not continuously
- Joint limits — mechanical hard stops, and hitting one at speed damages things
A trajectory that ignores these isn’t a trajectory, it’s a wish. → Motion Planning
Related
- Robot Control — the loops that drive these
- Electricity — power, current, and why motors are noisy
- Microcontrollers — PWM, timers, and what generates these signals
- Robotics map
Software engineering
Infrastructure & security
Computer science
Hardware & engineering