How to Choose Arm Joint Actuators for a Wheeled Humanoid
An arm joint actuator for a wheeled humanoid is the module of motor, speed reducer, encoder and bearings that drives each degree of freedom in the shoulder, elbow, wrist and torso. Selecting one is a trade-off between torque density, reflected inertia, precision, noise and collision tolerance. This guide walks through those trade-offs in the order they usually bite.
Why wheeled humanoids change the actuator question
A wheeled humanoid skips the hardest actuation problem in legged robots: knees and hips that absorb walking impacts at 2 to 4 times body weight, every step. That budget does not disappear. It moves up the body.
On a wheeled platform the arms ARE the product. The robot earns its keep by picking, placing, wiping, scanning and handing over objects in spaces built for people. That shifts the actuator requirements in four ways:
- Manipulation precision replaces gait stability. End-effector accuracy is set by the sum of joint errors along a 7-DoF arm. Backlash anywhere in the chain multiplies through the kinematics.
- The environment is full of humans. A warehouse aisle, a hotel corridor, a lab bench. Noise and safe physical interaction stop being nice-to-haves.
- Collisions are routine, not exceptional. A wheeled base cannot step away. Arms bump into carts, shelves and people at low speed, constantly.
- The distal mass problem dominates the mass budget. Every kilogram in a wrist actuator demands stronger elbow and shoulder actuators to carry it, and a heavier base to keep the robot from tipping. Mass at the wrist compounds three times before it reaches the wheels.
That fourth point deserves its own section, because it drives more design decisions than any datasheet number.
What is the distal mass problem?
The distal mass problem is the compounding penalty of putting heavy actuators far from the torso: distal joint mass must be carried by every proximal joint, and the whole arm’s mass must be counterweighted by the base for tipping stability.
Designers respond by concentrating mass near the torso and keeping limb inertia low. For the actuator selection this means the wrist and elbow reward small diameter, low weight and high torque density more than any other spec. A speed reducer that delivers 75 Nm from 0.8 kg in a 65 mm diameter changes what an arm can look like. That is the class of number to compare, not gear ratio alone.
The six criteria that matter
1. Torque density (Nm per kg)
Arms live and die by it, because of the compounding above. Compare complete-joint numbers where possible: motor plus reducer plus encoder plus bearings. A reducer that needs an oversized safety margin against shock loads drags the whole joint down.
2. Reflected inertia and backdrivability
Reflected inertia is the motor’s rotor inertia multiplied by the gear ratio squared. It is what a person feels when they push against the arm, and what the arm feels when it touches a surface. High reduction ratios (100:1 and up) make joints stiff-feeling and hard to backdrive. Quasi-direct-drive designs go the other way: ratios of 6:1 to 15:1, big motors, excellent force transparency, at the cost of heavy motors and high continuous currents. For wheeled humanoids doing contact-rich tasks near people, low reflected inertia is a safety feature as much as a control feature.
3. Backlash and lost motion
Backlash is play between transmission elements when the direction reverses. In a 7-DoF arm the end effector sees the sum of all joint backlash, scaled by link lengths. For repeatable pick-and-place, sub-arcmin lost motion per joint is the realistic target. Also check how backlash grows over life: gear teeth wear, and a joint that ships at 1 arcmin can drift to 3 arcmin or more. Rolling-contact transmissions do not have teeth to wear, which is why they hold their day-one backlash over the service life.
4. Noise
A humanoid working next to people is heard before it is seen. Gear meshing is the dominant noise source in most joints. Planetary stages typically exceed 60 dB; strain wave gears do better; transmissions without tooth engagement do better still. Specify dB at working speed and load, not at idle. More on this in our guide to quiet gearboxes for robot joints.
5. Collision tolerance and overtorque protection
Arms on mobile bases collide. The failure you are protecting against is a torque spike snapping a flexspline or brinelling a bearing. Options: mechanical clutches (bulky), software torque limits (latency-bound), or transmissions with a built-in slip mechanism. The Archimedes Drive handles overtorque through micro-slip: temporary sliding of the traction contact without meaningful damage, which caps the transmitted torque without an added component. That removes the need to oversize the reducer for worst-case impacts, which feeds straight back into criterion 1.
6. Wear and maintenance over life
Service robots are judged on fleet uptime. Ask any vendor: what does backlash look like after 10,000 hours? What lubrication schedule does the transmission need, and can field staff do it? Transmissions with sliding tooth contact wear differently than rolling contact. This is where lifetime cost diverges from purchase price.
The four transmission architectures, honestly
| Planetary / QDD | Strain wave (harmonic) | Cycloidal | Traction (Archimedes Drive) | |
|---|---|---|---|---|
| Typical ratio | 6:1 to 15:1 (QDD) | 50:1 to 160:1 | 30:1 to 200:1 | high single-stage ratios |
| Backlash | moderate, grows with wear | near zero when new, wear-limited | low | 0.2 arcmin max lost motion, zero backlash for life |
| Efficiency | up to ~95% | ~70-80% | ~80-90% | 85% |
| Backdrivability | excellent (QDD) | poor | moderate | good, low reflected inertia |
| Shock response | motor absorbs (QDD) | flexspline is the fuse | robust | micro-slip: temporary sliding, no meaningful damage |
| Noise source | gear mesh | tooth engagement | pin/roller engagement | rolling contact, no teeth |
| Weak point | motor mass (QDD) | flexspline fatigue | weight at small sizes | newer technology, fewer fielded years |
Each column wins somewhere. QDD wins raw force transparency. Strain wave wins packaging at high ratios and has decades of fielded history. Cycloidal wins ruggedness in heavy industrial joints. Traction drives win the combination this article is about: zero backlash for life, silence, collision tolerance and arm-friendly mass, at 85% efficiency. Read more about how a traction drive speed reducer works.
FAQ
What gear ratio does a wheeled humanoid arm joint need?
Shoulder and elbow joints typically run high ratios for holding torque against gravity. Wrists trade toward lower ratios for speed and force sensitivity. There is no single number; derive it from worst-case static torque with the payload at full reach, then check dynamic cases.
Is zero backlash worth it on every joint?
On arms, the joints closest to the end effector benefit most, because their errors are least attenuated. On a torso pan joint, moderate backlash may be acceptable.
How loud is too loud for a service robot?
Indoor human environments generally want the robot under normal conversation level (about 60 dB) at 1 m. The transmission is usually the first place to look.
Why not just use the actuators legged humanoids use?
You can, but you inherit their compromises: shock-rated mass you do not need, and noise and backlash trade-offs made for locomotion rather than manipulation.