Humanoid robot actuator layout, joint by joint

By Tolerance Labs engineering team · Updated 18 October 2026

A humanoid typically has 20 to 30 joints before its hands, and they do not all need the same actuator. This guide covers which joints carry the most torque, which reducer each usually gets, and where the mass should go.

A typical degree-of-freedom layout

Most full-body humanoids share a similar skeleton, though many designs add or drop a joint. Per side: a hip with 3 DoF (yaw, roll and pitch, usually meeting near one point so the hip acts like a ball joint), a knee with 1 (pitch), an ankle with 2 (pitch and roll), a shoulder with 3, an elbow with 1 and a wrist with 1 to 3. Once in the body: a waist with 1 to 3, yaw being the most common, and a neck with 1 or 2 for pointing cameras.

That gives 12 leg joints, 1 to 3 at the waist, 10 to 14 in the arms and 1 or 2 at the neck: roughly 24 to 31 before the hands. Each hand adds its own actuators; see how robot fingers are actuated.

Which joints carry the most torque

The knee is usually the highest-torque joint in the robot: in a squat, on stairs and when rising from a crouch, it holds the body with the foot well ahead of or behind it. Hip pitch swings the leg and holds the torso upright, so push-off, stairs and leaning forward load it heavily. Ankle pitch carries the body weight on the lever from the ankle to the center of pressure under the foot, which moves toward the toes at push-off.

Hip roll comes next. In single support the center of mass sits to the side of the stance hip, and hip roll holds it there on every step. Hip yaw and ankle roll carry less. Above the waist, shoulder pitch sees the most torque, because it carries the whole arm at full reach; the elbow, wrist and neck follow. The guide on sizing robot arm joints works through those numbers.

Gearing joint by joint

Legs and arms ask different things of a reducer. Leg joints hit the ground on every step, so they need to tolerate impact and, ideally, backdrive so the controller can run the leg as a spring and damper. Small and medium humanoids often use quasi-direct-drive joints at about 6:1 to 10:1 there. Heavier robots outgrow one low-ratio stage and move to two-stage planetary or cycloidal reducers, harmonic drives or screw-driven linear actuators, at a cost in backdrivability and reflected inertia.

Arm joints carry lighter loads at longer reach, hold poses for long periods and must fit in a slim limb, so harmonic drives at 50:1 to 100:1 or more are common at the shoulder and elbow. The table lists typical patterns, not rules; actual torques depend on robot mass, leg length and gait. The trade-offs behind each choice are in QDD vs high-ratio joints and harmonic vs planetary vs cycloidal reducers.

Ankle actuation: at the joint or up the leg

The ankle needs high torque, but it sits at the end of the leg, where every gram swings with each step. Designers use three layouts.

• Rotary actuators at the ankle. One module per axis. Simple kinematics and full torque at every angle, but the heaviest lower leg.
• Two linear actuators on the shank with a parallel linkage. Screw-driven actuators mounted high on the shank push rods down to the foot. Moving both rods the same way pitches the foot; moving them in opposite directions rolls it. The motors sit near the knee and share the pitch load, so each can be smaller.
• Rotary actuators near the knee with push rods. The same idea with rotary joints.

The linkage layouts have costs. The ratio from actuator to joint changes with angle, so available torque falls toward the ends of the range. Pitch and roll are coupled, so the controller needs the linkage kinematics. Range of motion is usually smaller, and how a screw takes impacts depends on its type and lead. Check the force rating at the worst linkage angle, as the micro linear servo guide does for fingers.

Mass budget: why distal mass matters

Actuators are often the largest single item in a humanoid’s mass budget, and where they sit matters as much as what they weigh. Mass far from a joint adds gravity torque in proportion to its distance and inertia in proportion to the distance squared.

Example: a leg 0.8 m long swings about the hip. A 1 kg actuator at the ankle adds 1 × 0.8² = 0.64 kg·m² of inertia about the hip. The same actuator at the knee, 0.4 m from the hip, adds 1 × 0.4² = 0.16 kg·m², a quarter as much. The hip needs four times the torque to swing the ankle-mounted mass at the same angular acceleration, and the foot lands with more effective mass.

The same logic runs up the arm, and the effect compounds: heavier distal joints raise the torque, and so the mass, of every joint closer to the body. Put the heaviest actuators near the hips and torso, use linkages, belts or tendons to move mass inward, and use compact joints under 500 g at the wrists and neck where the torque allows. Expect to iterate the layout two or three times.

Worked example: scaling knee torque with robot mass

This applies the method in sizing joint torque for a legged robot to a biped. The figures are illustrative.

• A 30 kg humanoid squats on both feet, so each leg carries about 15 kg. Force on each foot: 15 × 9.81 ≈ 147 N.
• In a deep squat the foot sits about 0.12 m horizontally from the knee. Static knee torque ≈ 147 N × 0.12 m ≈ 17.7 N·m.
• Rising quickly, stepping down stairs and landing raise the ground force. With a dynamic factor of 2.5, peak knee torque ≈ 2.5 × 17.7 ≈ 44 N·m.

Now double the robot to 60 kg. With the same proportions and density, mass grows with the cube of height, so the legs are about ∛2 ≈ 1.26 times longer, and the knee offset grows to 0.12 × 1.26 ≈ 0.151 m.

• Force on each foot: 30 × 9.81 ≈ 294 N.
• Static knee torque: 294 N × 0.151 m ≈ 44.5 N·m.
• Peak with the same factor: 2.5 × 44.5 ≈ 111 N·m.

Doubling the mass raised the peak knee torque about 2.5 times (2 × 1.26), because the lever grew too. In single support, such as one step on a stair, the stance knee carries the whole body, which doubles the static figure again. That is why adult-size humanoids often need joints of 100 N·m or more at the knee and hip. The joint torque calculator runs the leg case with your own numbers.

Checklist

• List every joint with its axis, range of motion and worst-case task.
• Estimate leg torques from mass, pose and a dynamic factor; redo them after each mass update.
• Use QDD in the legs where the torque allows; move to higher ratios or linear actuators only where it does not.
• Use harmonic or other high-ratio joints in the arms, with brakes where the arm must hold without power.
• Decide the ankle layout early: it sets shank mass, range of motion and controller kinematics.
• Check standing heat: RMS torque against continuous rating, as in how to size a robot joint motor.
• Standardize on one bus voltage and three or four module sizes; see power supplies and batteries for robot joints.

Browse humanoid and legged robot joints, quasi-direct-drive joints and harmonic joint modules, or see complete humanoid robots.