Selecting actuators for a quadruped is often reduced to "calculate the torque, then buy the biggest one". After integration, the usual surprises are a drifting gait, excessive positional play and thermal derating after a few minutes. The problem is rarely the torque figure itself.
Start with the real load torque
Joint torque comes from three sources:
- Static load: the moment produced by body weight on that lever arm. Worst case occurs at specific postures, not standing still.
- Inertial load: angular inertia during swing-leg acceleration and deceleration. At high gait frequencies this can exceed the static load.
- Friction and coupling: gearbox efficiency, bearing damping and reaction forces from adjacent joints.
Sizing from a static standing pose alone underestimates dynamic demand. Estimate from the intended gait's angular acceleration profile, then apply a margin covering friction and ageing.
Coreless, iron-core and brushless
- **Iron-core brushed**: cost-effective with good torque density and high starting torque, suitable for small and medium joints. Drawbacks are cogging at low speed and heat under sustained high current.
- **Coreless**: low inertia, fast response, no cogging — good where dynamic response dominates. Torque is limited at a given volume and heat dissipation is weaker.
- **Brushless**: clear advantages in lifetime and efficiency, but needs a driver, closed-loop feedback and higher integration effort.
For quadrupeds, iron-core brushed remains the mainstream choice in the 20–35 kg·cm range: the best balance of cost, torque and availability. Moving to brushless should follow the platform's lifecycle and maintenance requirements, not a single torque figure.
Backlash ruins gaits faster than torque does
Gearbox backlash creates dead travel when a joint reverses direction. On a quadruped this shows up as foot placement error, a slight sway while standing, and control-loop oscillation.
Treat backlash as a hard specification, and note that:
- Published figures are typically no-load and new-unit values; they grow with wear.
- Harmonic and planetary reduction differ substantially in backlash, and in cost.
- Closed-loop control mitigates but cannot eliminate mechanical play; where the sensor sits relative to the gearbox matters.
Consistency is the hidden barrier in volume builds
Torque constant, no-load current and dead-band width vary between batches. If the four legs have dissimilar actuator characteristics, the controller needs per-joint tuning or the gait becomes asymmetric.
For volume procurement, request batch consistency data rather than a single sample datasheet.
Acceptance checks
- Measure stall torque and the temperature-rise curve; confirm the derating under continuous duty.
- Measure backlash through a direction reversal with an angular indicator, not by eye.
- Run the target gait continuously for at least twenty minutes and watch for missed steps or protection events.
- Sample several units from the same batch and compare parameter spread.
Conclusion
Size torque from dynamic load first, then screen on backlash and consistency, and only then compare brands and price. Torque determines whether the robot can move at all; backlash and consistency determine whether it moves well.