How to build a million actuators

Everyone is talking about reindustrialization, and you might catch me among them. Hopes are high that robotics will fill the workforce gap, bring down the cost of making things, and usher in a new golden age of industrial production in the West. But who will build the robots, and how?

In The World Is Flat, Thomas Friedman argued that globalization was removing geographical barriers to competition. Today, I can buy actuators or complete robots from China with a few clicks. Making the same part in California or Europe is another story. Decades of specialization, investment, and industrial policy have built supplier ecosystems that are hard to replicate on either side of the Atlantic (Apple in China tells this story well). Could we design products and factories differently, using automation to make up for the suppliers and industrial know-how we no longer have?

I have been wondering what it would take to design a fully autonomous production line for robot actuators. Let's try to do just that. Through the teardown of a real robot actuator, I'll dive into its components, the key design decisions involved in building one, and how those decisions affect the way it could be manufactured in a fully autonomous factory.

My name is Matthieu. I am a mechanical engineer from ETH Zurich that spent his early career building sensors and robots systems.

Why care about actuators?

An actuator is the unit at each joint of a robot that turns electrical power into controlled motion. Electric actuators now power some of the most advanced robots in the world. Take Boston Dynamics' Atlas, which went from hydraulics to an all-electric design.

A rotary actuator typically provides one degree of freedom. Atlas has 56. Elon Musk has talked about producing 20,000 Optimus robots per week. Actuators account for an estimated 40–60% of a humanoid robot's hardware cost. If robotics startups deliver on their promises to their investors, demand for actuators is set to balloon.

This makes actuators an interesting case for reindustrialization: mechanically complex, relatively expensive, and potentially needed in enormous volumes. Could fully automated, flexible manufacturing make it economical to produce them in the West?

In the coming weeks, I'll explore the key components of an actuator—the motor, gearbox, and control board—and how each could be produced in a simulated manufacturing plant.

To keep complexity in check, we'll anchor our research in Unitree's GO-M8010-6, the hip actuator of its Go2 robot dog.

Enjoy!
mgsa

Where it lives

The Go2 has twelve of these modules, three per leg: one swings the leg out from the body, one swings it forward and back, one bends the knee. Below, the robot is drawn from Unitree's own published description of it, with the twelve modules marked and the three of the front-left leg in orange. Press play to see what each one does during a stride. The knee module drives the calf through a rod, which is why that joint is rated at twice the torque and half the speed of the other two.

Model: Unitree Go2 robot description from the unitree_ros package (BSD-3 licence), meshes simplified; joint ranges and ratings from the same file. The stride is a drawn gait, not a recording.

The module, exploded

Drag the slider to pull the GO-M8010-6 apart along its axis. One label per chapter.

Model: Unitree customer STEP (2022) for the outer parts; internals reconstructed, dimensions estimated.

Chapters

1. 1
   The brushless motor
2. 2
   The planetary gearboxupcoming
3. 3
   The encoderupcoming
4. 4
   The driver boardupcoming
5. 5
   Housing, bearings and heatupcoming
6. 6
   Assembly and end-of-line testupcoming
7. 7
   The lineupcoming

Also on this site