I've spent over two decades converting rotational energy into linear motion, and if there is one thing that keeps me up at night, it's the "demo video" culture in modern robotics. We see robots dancing and doing backflips in padded labs, but as an engineer, I'm looking at the ankles. I'm looking at the thermal footprint. I'm wondering how many operating hours that gearbox has left before it shears a tooth.

If a robot can't survive a double-shift in a "Dull, Dirty, or Dangerous" environment without destroying a joint, it isn't a revolution — it's a toy. The truth that most software-first startups want to ignore is that actuators are the primary engineering driver for humanoid robots. Not the AI. Not the software stack. Not the pitch deck. The actuator — the component that converts electrical energy into physical movement — determines whether a robot works or falls apart.

If you're new to robotics, you might be wondering what an actuator even is and why it matters so much. This article will explain everything from the ground up. If you already know the basics, skip ahead to the engineering trade-offs — that's where it gets interesting.

What Is an Actuator? The Muscle of Every Robot

An actuator is any device that converts energy into physical motion. In the human body, muscles are actuators — they convert chemical energy from food into mechanical force that moves your bones. In a robot, actuators convert electrical energy from a battery into mechanical force that moves the robot's limbs.

Every moving part on a humanoid robot — every joint in the legs, arms, hands, torso, and neck — requires at least one actuator. Without actuators, a robot is just a metal skeleton. The actuator is what makes it move, lift, grip, walk, and balance. It is, without exaggeration, the most critical component on the entire machine.

There are two fundamental types of actuators used in humanoid robotics: rotary actuators and linear actuators. Understanding the difference between these two — and knowing where each one belongs on a robot — is the single most important engineering decision in humanoid design. Get it right and you have a machine that can work an 8-hour shift. Get it wrong and you have a machine that destroys itself in a week.

Rotary Actuators: The Spinning Joints

A rotary actuator produces rotational motion — it spins. Think of an electric motor turning a wheel. The output is torque (rotational force) delivered around a central axis. When you open a door, the hinge is the axis and your arm is applying torque to rotate the door around that hinge. A rotary actuator does the same thing, but with an electric motor instead of your arm.

In a humanoid robot, rotary actuators are used at joints that need a wide, continuous range of rotational motion. The most obvious example is the shoulder — your shoulder joint allows your arm to swing forward and backward, side to side, and rotate inward and outward. A rotary actuator sitting directly at the shoulder pivot can replicate all of this movement.

How They Work

Most rotary actuators in humanoid robots consist of two components working together: an electric motor and a gear reducer. The electric motor — typically a Brushless DC (BLDC) motor — spins very fast but produces relatively low torque. The gear reducer takes that high-speed, low-torque rotation and converts it into low-speed, high-torque rotation. Think of it like a bicycle's gears — when you shift to a lower gear, you pedal faster but the wheel turns slower with more force. Same principle.

The most common type of gear reducer in high-end humanoid robotics is called a harmonic drive (also known as a strain wave gear). A harmonic drive is an elegant piece of engineering — it uses a flexible steel ring (called a flexspline) that deforms inside a rigid outer ring (the circular spline) to achieve gear reduction ratios of 50:1 to 160:1 in a package the size of a hockey puck. The result is a compact, lightweight, zero-backlash rotary actuator that can produce impressive torque in a very small volume.

Where They Excel

Rotary actuators are ideal for joints where you need smooth, continuous rotation through a large arc. On a humanoid robot, this means the hips (which need to swing the entire leg through a wide arc), the shoulders (which need even greater range of motion), the elbows, and the wrists. These joints need to rotate freely and continuously, and a rotary actuator sitting directly on the pivot is the most mechanically efficient way to achieve that.

The compact form factor of harmonic drives is a major advantage here. A rotary actuator built around a harmonic drive can fit inside the body of a robot arm without creating bulges or protrusions that would interfere with movement or catch on objects in the environment. This slim profile is critical for arms and wrists that need to reach into tight spaces.

Where They Fail

Here's where it gets real. Harmonic drives have a critical weakness: they are terrible at absorbing shock loads.

Remember that flexspline — the thin, flexible steel ring at the heart of the harmonic drive? It's what gives the mechanism its zero-backlash precision. But thin, flexible steel is also fragile under sudden impact. When a 90 kg robot lands from a jump, or stumbles on an uneven surface, or has its foot catch on an obstacle, the instantaneous spike load through the leg joints can exceed the harmonic drive's rated torque by 5 to 10 times. That shock energy concentrates on a handful of thin gear teeth.

The result is fatigue failure. Not after one impact — after thousands. Each shock creates microscopic damage in the flexspline. After 10,000 steps, 50,000 steps, 100,000 steps, those micro-cracks propagate until the flexspline fractures. The harmonic drive fails, the joint collapses, and you've destroyed a $2,000 component. In a factory running two shifts, a robot takes roughly 80,000 steps per day. A harmonic drive at the knee of a walking humanoid is operating in a regime it was never designed for.

This is why you don't see harmonic drives in the legs of any serious humanoid robot. Not in the Tesla Optimus. Not in the Boston Dynamics Atlas. Not in any machine designed to actually walk on concrete floors for hours at a time. The physics simply doesn't allow it.

Linear Actuators: The Pushing Joints

A linear actuator produces straight-line motion — it pushes or pulls. Instead of spinning, it extends and retracts along a single axis, like a piston. Think of your leg when you stand up from a chair: your quadriceps muscle contracts (shortens) in a straight line, pulling on a tendon that's attached to your kneecap, which levers your lower leg into extension. That's a biological linear actuator.

In a humanoid robot, linear actuators are used at joints that must absorb heavy impact loads and deliver high force in a compact, protected package. The most critical application is the knee — the joint that bears the full weight of the robot on every single step and absorbs the brutal shock of foot-to-ground contact thousands of times per hour.

How They Work

An electric linear actuator in a humanoid robot typically consists of a BLDC motor connected to a screw mechanism. The motor spins the screw, and a nut riding on the screw threads converts that rotation into linear (straight-line) motion. As the screw turns, the nut moves forward or backward along its length. Attach the nut to a lever arm on the robot's leg, and you've created a linear actuator that can push or pull with enormous force.

There are two main types of screw mechanisms used in humanoid robotics:

Ball screws use steel balls rolling in grooves between the screw and the nut. They're efficient and widely available, but the balls make point contact with the screw — meaning the load is concentrated on tiny spots. Under high shock loads, this point contact creates a phenomenon called "Brinelling" — small dents form in the raceway where the balls impact. Over thousands of cycles, these dents grow until the screw mechanism fails.

Planetary roller screws are the premium choice and the technology that dominates in high-performance humanoid legs. Instead of balls, they use threaded rollers that make line contact with the screw. Imagine the difference between pressing a marble into clay (point contact — leaves a deep dent) versus pressing a pencil into clay (line contact — distributes the force across a much larger area). Planetary roller screws have 10 to 15 times more contact surface area than ball screws. This means they distribute shock loads across dozens of threaded contact points simultaneously, rather than concentrating them on a few steel balls. The result is dramatically superior shock absorption and fatigue life — exactly what a robot knee needs.

Where They Excel

Linear actuators dominate in the lower extremities of humanoid robots for two reasons that I have been preaching to engineers for over 20 years.

First: shock absorption. Every step a humanoid takes generates an impulse load at the knee. For a 70 kg robot walking at a normal pace, peak knee forces reach 2 to 3 times body weight — roughly 1,400 to 2,100 Newtons — delivered in less than 50 milliseconds. At a jog, those forces spike to 5 to 7 times body weight. A planetary roller screw absorbs this energy across its entire threaded surface area. A harmonic drive concentrates it on a few thin teeth. Over 80,000 steps per day, the roller screw survives. The harmonic drive doesn't.

Second: slender profiles. A linear actuator can be mounted parallel to the structural members of the leg, running alongside the bone-like struts the way a human calf muscle runs alongside the tibia. This keeps the leg slim enough to fit through doorways, navigate between furniture, and operate in human-scale environments. A large rotary motor mounted directly at the knee pivot creates a bulky, protruding joint that limits range of motion and catches on obstacles.

There's a reason that the human body uses a linear tendon system in the legs rather than a rotary joint mechanism. Three hundred million years of evolutionary optimization arrived at this architecture because it works. The quadriceps muscle pulls linearly on the patellar tendon, which wraps over the kneecap and converts that linear pull into rotation of the lower leg. The most successful humanoid robots — Tesla Optimus, Boston Dynamics Atlas, Figure 02 — all mimic this biomechanical architecture with linear actuators driving the knees and ankles through lever arms, exactly the way your muscles drive your bones through tendons.

Where They Have Limitations

Linear actuators are not ideal for every joint on a robot. Shoulders and hips need multi-axis rotation through wide arcs — a linear actuator can only push or pull along one axis. You can convert linear motion to rotation through a linkage mechanism, but at the shoulder, where you need three independent axes of rotation with 180+ degrees of range, the mechanical complexity of a linear-to-rotary linkage system becomes impractical. Rotary actuators are the right choice for these joints.

Linear actuators also have a mechanical disadvantage issue. Because a linear actuator is mounted offset from the joint pivot and pushes against a short lever arm, it must generate much greater raw force than the equivalent torque from a direct-drive rotary motor. For a typical humanoid knee, a linear actuator mounted 50mm from the pivot must produce over 7,000 Newtons of force to match 353 Nm of torque from a rotary motor — roughly 1,600 pounds of push force. Planetary roller screws can deliver this force, but it means the linear actuator is working harder for every Newton-metre of joint torque. The trade-off is worth it for the shock absorption and fatigue life advantages, but it's a real engineering constraint.

The "Walking Problem": Why This Matters More Than AI

A humanoid robot takes roughly 5,000 steps per hour at a normal walking pace. Each step sends a shock of 2 to 3 times body weight through the leg actuators. Over an 8-hour shift, that's 40,000 impact cycles. Over a year of two-shift operation, that's roughly 20 million impact cycles.

No other component on the robot faces this kind of mechanical punishment. The cameras don't get hit. The computer doesn't absorb shock. The batteries sit in a protected compartment. But the leg actuators — specifically the knee and ankle joints — must survive relentless, repeated, high-energy impacts while simultaneously producing the precise torque needed to keep a 70-to-90 kg machine balanced on two feet.

This is why actuators are the most important part of the machine: they are the interface between digital intelligence and a chaotic, high-impact physical world. The best AI in the world is useless if the hardware under it shatters at step 50,000. And step 50,000 arrives in less than a week of full-time operation.

If an actuator is mechanically self-locking — as most industrial lead screws are — the gearbox is forced to absorb 100% of the shock energy instantaneously, leading to accelerated wear and eventual shear failure. This is why the choice of screw mechanism (ball screw vs. planetary roller screw) isn't an academic debate. It's the difference between a robot that works for five years and a robot that works for five weeks.

The Actuator Count: How Many Does It Take?

To mimic human mobility, the industry is converging on a specific count of degrees of freedom (DoF). Each degree of freedom — each independent axis of movement — requires at least one actuator. Here's how it breaks down:

The Legs (12-14 Actuators): Each leg requires 6 to 7 actuators. The hip needs 3 actuators for flexion/extension, abduction/adduction, and internal/external rotation. The knee needs 1 actuator for flexion/extension. The ankle needs 2 to 3 actuators for dorsiflexion/plantarflexion and inversion/eversion. The hip actuators are typically rotary (wide range of motion, moderate shock loads). The knee and ankle actuators are typically linear (narrow range of motion, extreme shock loads).

The Arms (14-16 Actuators): Each arm requires 7 to 8 actuators. The shoulder needs 3 actuators for the same three axes as the hip. The elbow needs 1 to 2 actuators for flexion and pronation/supination. The wrist needs 2 to 3 actuators for flexion/extension, radial/ulnar deviation, and pronation/supination. All arm actuators are typically rotary — the shock loads are manageable and the range of motion requirements favour continuous rotation.

The Hands (10-20 Actuators per hand): Dexterous manipulation requires micro actuators in each finger. Tesla's Optimus Gen 3 uses 25 actuators per hand with 22 degrees of freedom — a tendon-driven system where miniature motors in the forearm pull cables that flex the fingers. This is the most actuator-dense subsystem on the entire robot, and it's where the frontier of humanoid engineering currently sits.

The Torso and Neck (3-5 Actuators): The torso needs 1 to 2 actuators for rotation and lateral bending. The neck needs 2 to 3 actuators for looking up/down, left/right, and tilting.

The total comes to 40 to 55 actuators per robot for a basic humanoid, and up to 70 or more for one with fully dexterous hands. At this actuator count, every gram of mass matters. A 200-gram error at the ankle — choosing an actuator that's slightly too heavy — compounds into a 1.3 kg penalty at the hip once you upsize the hip motors, structural members, and battery to carry the extra weight at the extremity. This is called the mass penalty spiral, and it is the reason actuator selection dominates humanoid engineering decisions.

The Architecture That's Emerging: Linear Below, Rotary Above

Every serious humanoid robot company has converged on the same architectural split, and it's the one that biology figured out hundreds of millions of years ago:

Linear actuators for the lower extremities. Knees and ankles use planetary roller screw linear actuators mounted parallel to the leg structure, pushing against short lever arms to generate joint torque. This provides the shock absorption, fatigue life, and slender profile needed for bipedal locomotion on real-world surfaces.

Rotary actuators for the upper extremities. Shoulders, elbows, and wrists use compact rotary actuators with harmonic drives or cycloidal reducers. The shock loads in the upper body are manageable (you're not slamming your arms into the ground 5,000 times per hour), and the range-of-motion requirements favour continuous rotation.

Micro linear actuators for the hands. Finger manipulation uses tendon-driven systems where miniature linear actuators in the forearm pull cables that flex and extend each finger joint. This relocates the motor mass away from the fingertips (reducing the moment arm the wrist actuator must fight) while providing the precise, force-controlled movement needed for object manipulation.

Tesla Optimus uses this architecture. Boston Dynamics Atlas uses this architecture. Figure AI uses this architecture. They all arrived at the same conclusion independently because the physics demands it. There is no alternative architecture that survives the mechanical realities of bipedal locomotion at these mass and force requirements.

The Cost Reality

Actuators account for 50 to 70 percent of a humanoid robot's bill of materials. For a robot with 50 actuators — some of which are precision planetary roller screw assemblies and others are compact harmonic drive modules — the actuator subsystem alone can cost $15,000 to $40,000 at current production volumes.

This is why the actuator supply chain is the most strategically important supply chain in humanoid robotics. It's why LG just launched a dedicated actuator brand (LG Actuator AXIUM). It's why Hyundai Mobis is manufacturing actuators for Boston Dynamics Atlas. It's why Samsung Electro-Mechanics is evaluating entry into the market. And it's why Tesla is manufacturing its own actuators in-house — the same vertical integration strategy they used with battery cells.

The company that cracks high-volume manufacturing of reliable, affordable actuators — particularly roller screw linear actuators below $100 per unit — will own the humanoid robotics market. Everything else is software. And software runs on hardware that somebody has to build.

The Bottom Line

The winner of the humanoid race won't have the best AI. They will have the best actuators — the highest continuous torque ratings, the longest fatigue life under shock loading, and the lowest cost per Newton-metre of output force. The AI will catch up. Models will improve. Software will be patched. But a mechanically deficient actuator that shatters at cycle 50,000 cannot be fixed with a firmware update.

The robot starts with the actuator. It always has.

For a deep dive into the math behind these engineering decisions, see our Interactive Rotary vs. Linear Joint Force Calculator. For a comprehensive primer on actuator technology in humanoid robotics, read our Complete Guide to Humanoid Robot Actuators.