← All articles

How to choose a brushless motor for a robot arm and for a mobile cart: ten steps from a column in The Robot Report and what to ask a supplier based on them

Topics: Robots, Industry

A disassembled robot joint on a workbench: a brushless motor with copper windings, a gearbox, a control board and a position sensor

The short answer: according to a column in The Robot Report, a motor for a robot is chosen not by the voltage and power in a catalogue but from the task: what torque (rotational force) is needed, at what speed, how often, with what accelerations and decelerations, and how much heat will have to be removed. A robot arm joint and a cart wheel usually need different things: for the joint, smoothness at low speed and position accuracy matter; for the wheel, battery efficiency, traction and braking matter more. Below: the main points of the column in plain words, four formulas for a first estimate and the questions worth asking a robot supplier based on them.

Where this comes from

The column "How to choose a BLDC motor for robotic arms and mobile robots" was published in The Robot Report on 18 October 2026; the author is Faisal Mahmood. In the bio under the column he is presented as a specialist in digital marketing and tech content. In the text he refers to reference material from chip makers (STMicroelectronics, TI, Microchip) and, as a commercial example, to the BLDC.com controller catalogue. I read the column in full. It is an overview of approaches, not test results, so I retell it as the author's advice.

What BLDC is and why the motor alone is not what matters

BLDC (brushless DC) is a brushless direct-current motor: the rotor carries permanent magnets, the stator windings are switched by electronics (an inverter), and there are no brushes. The author notes that commercial texts often treat BLDC and PMSM (permanent magnet synchronous motor) as almost the same thing, because both are three-phase, use magnets and are electronically controlled. But how a particular motor behaves is determined by the shape of its back EMF (the voltage that a rotating motor itself induces in its windings), the winding design, the controller algorithm and the way the rotor position is sensed.

The main idea of the column: the actuator is a system. The result is influenced by the motor, inverter, current sensors, rotor position sensor, gearbox, brake, output sensor on the joint, heat path, communications interface and the robot's software further up the chain.

Where to start: with the load

The author suggests first establishing what the motor has to do, and gives four formulas for a first estimate.

  1. Joint torque: τ_joint = (τ_motor × i × η_gear) − τ_losses. Here τ is torque, i is the gear ratio, η is the gearbox efficiency. In words: the gearbox multiplies the motor torque by the gear ratio, part of the torque is lost in the gearbox itself (its efficiency accounts for this), and other losses are subtracted separately.
  2. Motor speed: ω_motor = ω_joint × i. In words: the motor turns faster than the joint by the gear ratio.
  3. Mechanical power: P_mech = τ × ω. Power equals torque multiplied by angular speed.
  4. Power from the battery: P_elec ≈ P_mech / η_total. The battery has to supply more than the shaft delivers, by the amount of all losses along the way.

A simple example with my own illustrative numbers, not from the column: if a joint needs 20 N·m, with a gear ratio of 50 and an efficiency of 0.8, then, ignoring other losses, the motor has to deliver 20 / (50 × 0.8) = 0.5 N·m. If the joint turns at 1 radian per second, the motor turns at 50 radians per second, about 480 revolutions per minute.

The author cautions that these formulas do not replace a dynamic model. But they do show what you need to know before choosing a controller. That is continuous and peak torque, how long the peak lasts, the torque for acceleration and for holding against gravity, the duty cycle, energy returned during braking, near-zero-speed operation, thermal resistance, the battery voltage range and the current limits of cables and connectors.

For a robot arm, add link inertia, payload, gearbox backlash and compliance, and where the position sensor sits. For a cart: mass, wheel radius, rolling resistance, slopes, acceleration, terrain, wheel slip and the drive layout.

The author singles out regenerative braking: during braking or downhill travel the motor can return energy to the actuator's power circuit (the DC bus). If the battery, controller and braking scheme are not ready for this, a controller that accelerates the robot without trouble can run into over-voltage during hard braking.

Three control methods

  • Six-step (trapezoidal). The inverter switches the windings through six states in turn according to rotor position. It is attractive where simplicity and cost matter and torque ripple (uneven force) is acceptable. It also fits when the motor mostly runs at moderate or high speed and precise low-speed torque is not the main concern. Rotor position can come from simple Hall sensors.
  • Sinusoidal. The current or voltage in the windings changes smoothly rather than in steps. Compared with simple six-step control, this can reduce ripple and noise, but the result depends on the motor, the sensors, current regulation and tuning.
  • Field-oriented (FOC, field-oriented control). The controller converts the three phase currents into two components tied to the rotor (the Clarke and Park transformations) and regulates them separately. One component (the q axis) is closely tied to torque; the current along the other (the d axis) can be used to control the magnetic field and, at high speeds, for field weakening. According to the author, FOC is usually chosen when smooth low-speed motion, fast torque response or accurate speed regulation is needed.

An important caveat from the author: FOC by itself does not make an actuator precise. If the position sensor is misaligned or the gearbox has large mechanical errors, even a very sophisticated algorithm cannot fix it.

Position sensors

  • Hall sensors give coarse rotor position. They are often enough for switching the windings and basic speed control, for example in inexpensive carts.
  • Incremental encoders give far more position points and are widely used in servo drives. Absolute encoders also know the position immediately after power-up, so the robot does not have to find "zero" first.
  • Sensorless control estimates position from the back EMF. It is cheaper and simpler to assemble, but at zero and very low speed the back EMF signal is hard to obtain, so starting from standstill and heavily loaded low-speed operation need a separate solution.

Where the sensor sits matters as much as accuracy. A sensor on the motor knows the motor's position, while at a joint behind a gearbox there is also backlash, torsional windup and transmission error. If these errors noticeably affect the task, a second sensor can be fitted at the joint output and the joint controlled by it. The author stresses that not every joint needs this: the mechanics of the particular transmission decide.

Arm and cart: what differs

Arm. A simple joint may be fine with a motor with Hall sensors and an external controller if it needs speed or basic positioning. A demanding joint is usually built with an encoder and FOC: a fast current loop runs in the drive itself, with speed and position loops above it. An integrated actuator combines some or all of the following: motor, gearbox, position sensor, drive electronics, brake, temperature sensing and communications interface; the set depends on the product. Such an actuator simplifies assembly but, according to the author, shifts the difficulty to heat removal, repair, firmware visibility and dependence on the actuator supplier.

Cart. Two independently driven wheels have to hold their commanded speeds accurately enough: stable motion, navigation and counting distance travelled from the wheels depend on this. During acceleration the battery voltage sags, and during regenerative braking the voltage in the actuator's power circuit can rise, so the controller has to be checked across the whole battery voltage range rather than at one nominal value. A drive inside a wheel or a cramped chassis may cool poorly.

How to read current on a controller datasheet

A datasheet may list continuous phase current, continuous bus current, peak phase current, peak duration, RMS current and a configurable limit. According to the author, these numbers are not interchangeable. The thermal limit of the motor, the limit of the inverter's power switches, connectors, cables and the battery can all be lower than the headline peak current figure. A controller with a large peak current is not necessarily suitable for a motor that needs that current continuously. It is worth comparing the voltage range, which current exactly is meant, how long it is allowed, compatibility with the sensors and the protection functions, not power alone.

Protection is not yet safety

The author lists what to check in a controller: protection against overcurrent, short circuit, over- and under-voltage, overtemperature, overspeed, encoder and Hall sensor failure, phase loss, stall, communications loss, voltage rise during regenerative braking and unintended restart, plus behaviour during emergency stop and safe torque off where it is needed.

But having an overcurrent shutdown or a temperature sensor does not, according to the author, make a safety function certified: it has to be designed, validated and integrated according to the applicable requirements. For industrial robots the author names ISO 10218-1:2025 (safety requirements for industrial robots), ISO 13849-1:2023 (safety-related parts of control systems, including software) and IEC 61800-5-2 (functional safety of adjustable speed electrical power drive systems).

Ten selection steps from the column

  1. Describe the load and motion in full: acceleration, deceleration, payload, terrain, gravity, duty cycle.
  2. Calculate continuous and peak torque at the motor shaft and at the joint or wheel.
  3. Establish the required speed, acceleration, operating range and regenerative braking conditions.
  4. Decide what you are controlling: speed, torque, position, stiffness (impedance) or traction.
  5. Choose sensors for this goal and for the mechanics of the transmission.
  6. Choose six-step, sinusoidal control or FOC.
  7. Check voltage, continuous and peak current, regeneration, thermal margin and battery behaviour.
  8. Choose how the actuator will exchange commands with the rest of the robot and its software. Examples from the column: CANopen with the CiA 402 profile and EtherCAT (ways of communicating between drives and the robot controller), ROS 2 via ros2_control (connecting the actuator to the robot's software).
  9. Define safe behaviour: response to faults, braking, communications loss, restart conditions.
  10. Validate the motor, drive, gearbox, sensors, brake and robot controller together, as one system.

What to ask a robot or actuator supplier based on this

This is for whoever buys a robot or actuator rather than designing it. The questions are asked by the buyer's project manager or engineer and answered by the supplier's or integrator's engineer. The questions are taken from the checklist at the end of the column.

  1. Before buying, ask in writing: what the actuator controls (speed, torque, position), whether Hall sensors are enough or an encoder is fitted, whether there is a sensor at the joint output, what margin for voltage, current and heat is built in, and what the actuator does on sensor failure, communications loss, braking and emergency stop.
  2. How to check: ask for a test report on the assembled actuator under your real load and duty cycle: with the gearbox and mechanism for an arm, with the actual wheel, mass and battery for a cart. According to the column, such tests include measuring supply voltage during braking, checking what the actuator does on communications loss and on sensor disconnection, checking emergency stop under controlled conditions and repeating the important tests after the motor and inverter have warmed up to operating temperature.
  3. If there is no report, or the tests were run on an unloaded motor on a bench: according to the author, such a bench test validates only part of the system. Do not sign acceptance until the supplier has run such tests under your load, with its own engineers and according to its own test programme, and you have received the report.

Summary

The column in The Robot Report reduces choosing a brushless motor for a robot to a simple idea: first the load and duty cycle, then the control method, sensors, current, heat, communications and safety, and finally testing the whole actuator as one system under real load. An arm joint and a cart wheel usually need different things, and the voltage and power figures in a catalogue do not give the answer. The column is an overview of approaches, not a calculation for a specific case, so it is best read as a list of questions rather than a replacement for an engineer's calculation.

See also another piece in the "Robots" section.

I work on AI agents and automation. If you would like to see my projects or discuss your own task, take a look at my portfolio.

Sources