How to Choose a Motor for a Quadruped Robot

1. Turn the gait into separate joint requirements

Record torque and speed together over time for each hip and knee axis. Base the model on leg geometry, body and payload mass, center of mass, link mass, foot clearance, joint travel and contact conditions. Include the intended walk, crouched hold and recovery step.

During stance, foot force changes with support contacts and load distribution. Dividing weight by four applies only to a particular static loading case. During swing, an unloaded foot still requires joint torque to accelerate the links and rotor.

The contact contribution is τ_contact = J(q)^TF. The Jacobian J maps joint motion to foot motion at posture q; it also maps contact force back to joint torque through its transpose. Use consistent coordinates and force signs. Add link-gravity, acceleration and friction torques without counting any load twice. See Northwestern University: the Jacobian force–torque relationship (external source).

A small calculation shows why posture matters

Illustrative assumptions, not product specifications: take a 100 N vertical foot force. For one joint, a perpendicular moment arm of 0.12 m produces a 12 N·m contact contribution; at 0.06 m, it produces 6 N·m. These are two assumed geometric cases, not a statement that deeper crouching always increases torque. Link gravity, acceleration and other joint axes are omitted from this simple check.

For a preliminary forward-driving estimate, define reduction N = motor speed / joint speed and efficiency η. Then τ_motor ≈ τ_joint / (Nη) and ω_motor = Nω_joint. With assumed N = 8, η = 0.90 and joint speed 4 rad/s, the 12 N·m point maps to about 1.67 N·m at 32 rad/s on the motor shaft. Its joint mechanical power is 48 W; the estimated motor mechanical input to the gearbox is 53.3 W. This is not battery power, and it omits acceleration of rotating components and detailed losses.

2. Size for the repeated cycle, not one successful step

Plot the required torque–speed samples against the supplier’s operating envelope at the intended voltage, temperature and cooling, and check whether each lies within it. Include both directions and braking. Maximum torque and maximum speed may occur under different conditions: check the required pair, then its permitted duration and recovery time.

For a first copper-loss screen, use the actual motor-current history: I_rms = √[Σ(I_i²Δt_i) / ΣΔt_i]. Keep the phase/line and peak/RMS conventions consistent with the winding resistance and torque constant. Resistive loss depends on current squared; OpenStax: electrical energy and resistive power (external source) explains the underlying principle.

Hypothetical one-second motor-current cycle: 2 A for 0.6 s, 5 A for 0.2 s and 1 A for 0.2 s gives I_rms ≈ 2.76 A, despite an arithmetic mean of 2.4 A. These are assumed calculation inputs, not SolisMotor specifications, a measured gait or allowable current limits.

RMS is a copper-loss screen. Long holds and overloads need a time-dependent thermal check that also considers iron, gearbox and driver losses. Record initial temperature, ambient conditions and the installed cooling path. Read continuous torque versus peak torque before comparing catalog figures.

3. Choose reduction and feedback around contact behavior

More reduction lowers the motor torque needed for a given forward-driven output load, while increasing motor speed for a given joint speed. It also changes the dynamics felt at the joint. For a simple rigid gear model, the rotor contribution reflected to the output is J_reflected = N²J_rotor; gearbox and link inertias must be added separately. Doubling N makes this rotor contribution four times larger if the rotor is unchanged. The convention and formula are documented by MIT Drake: reflected rotor inertia (external source).

Low ratios can help limit this inertia contribution, but may increase motor torque and current demand. Compare friction, compliance, backlash and contact response alongside the ratio. Use consistent test conditions when reading how QDD actuation works and what backdrivability means at the actuator output.

Motor-side feedback measures rotor motion; output-side feedback measures position after the transmission. Dual encoders alone do not measure torque. Ask for encoder accuracy, latency, update rate, startup behavior and calibration data; bit count only describes resolution.

The SolisMotor MG8108-9 product page is an integrated planetary-joint reference for an initial shortlist. Use the confirmation table below to request the missing operating and integration evidence. The hypothetical calculations above do not qualify this model for a quadruped.

4. Qualify a representative leg before scaling to the robot

Set acceptance criteria for trajectory tracking, temperature, contact response and faults before testing. Progress from restrained joint/leg fixtures to supported whole-body trials under the project’s test procedure.

1. Replay the working points. Use the intended links and payload representation. Record joint position, speed, measured or appropriately estimated torque, current and command timing together.
2. Repeat the contact cycle. Include stance, swing, push-off, touchdown and a credible recovery maneuver. Inspect load paths, bearings, fasteners, cable travel and feedback continuity.
3. Reach the relevant thermal condition. Test with the intended housing, mounting contact and cooling. Log motor and driver temperatures and any derating. A housing sensor is not automatically a winding-hotspot measurement.
4. Check system interactions. Verify communication timing, voltage at the drive during simultaneous joint loading, braking-energy handling and the defined response to sensor, communication or power loss.

Keep phase current and DC-bus current distinct. Their relationship depends on switching and operating state. Also check the power system’s ability to accept or dissipate returned energy during deceleration; a motor-only test leaves that system question unresolved.

Complete one confirmation sheet per joint axis

Keep the selection provisional if a required working point is unverified, overload recovery is unknown, or the rating’s cooling conditions differ from the installation. Record each gap and the calculation, supplier confirmation or test needed to close it. The SolisMotor robotic joint drive solutions overview describes application-selection and system-integration support.

Frequently asked questions

How much torque does a quadruped robot need?

There is no universal value. Calculate each joint from leg posture, contact forces, link dynamics and the intended gait. Robot weight alone does not capture moment arms, acceleration or changing support contacts.

Is peak torque enough to select a leg motor?

No. Check the simultaneous speed, allowable peak duration and repetition, repeated-load temperature and drive-current limits. Include crouched holds and recovery steps as well as normal walking.

Are QDD motors always the best choice?

No. Low-ratio geared joints can be useful for dynamic interaction, but the required torque, speed, mass, cooling, contact behavior and control performance must be evaluated together.

Can the same motor be used at every joint?

Possibly, if each axis meets its own operating and structural requirements. A common motor can simplify integration and spares, but may add unnecessary mass or leave one axis short of capability.

Does a high-resolution encoder guarantee accurate foot-force control?

No. Encoder accuracy and latency, torque estimation, transmission behavior, geometry calibration and controller response also affect the result. Validate foot-contact behavior on the assembled leg.