← All posts

Building the American Actuator Company

This is an edited version of an update I recently shared with our customers at the Saturn Robotics Corporation. I hope you find it useful in understanding what it takes to build an Actuator company in America.

10min read · Saturn Robotics

Tl;Dr: We built our own production machines to make motors because it is the only long term economical way to do so.

Building the American Actuator Company

About 6 months ago it became clear to me that my plan for long term economics of mass manufacturing as I had originally envisioned doesn't work out. Even with the unusually high prices of brushless motors and actuators, the cost of manufacturing in the US along with aggressive competition from overseas will mean you dip into negative margins very quickly.

For a time I thought copy pasting the same machines and automation systems used in Chinese factories, plus the closer market access together will be sufficient. However, even then you run the risk of falling into the red because Chinese manufacturers will always have pricing power and other advantages that you cannot replicate.

This would be an existential risk to the business and you'd only be able to compete in the much smaller markets that must buy domestic. The long lead times of these automation systems also meant you will remain slow to scale. Breakdown in machines you didn't design means you'll be flying around the world getting technical support. You can buy some of these machines locally or in places with better language support, but you end up paying exorbitant prices for the most common of machines like CNC mills.

The price at which you sell a mass manufactured product will include cost of raw materials, shipping, land/factory space, labor, amortized cost of electricity, insurance, consumables, depreciation schedule of all machinery used, downtime costs, front office costs, marketing expenses, cost of capital, and acts of god.

We spent the last 6 months aggressively tackling all of these save the acts of god. Primarily, we tackled 3 of the most expensive challenges: Labor, Cost of Machinery, and Front office costs. I'll start with the first 2 as they go hand in hand.

Labor and Cost of Machinery

Labor costs what it costs, and certainly if you want good work out of people you have to pay them well. The way you can have it cost less to you is by empowering workers with as much machinery and automation as possible. To increase worker productivity we need a lot of machines and machines are expensive. So we decided to build our own. Production can be broken into 3 categories: Machining, Assembly, and Testing.

Machining

V1 Mill got us quite far. V2 currently in fabrication will be bigger, beefier, and a lot more accurate. We started with building our own mill and lathe, which covers the bulk of machining. Simple enough but this is what took the most amount of time. We built V1 with off the shelf motion components and a heavy steel frame.

Not counting the mental toll of the anguish of figuring out why your machine is cutting ovals, the final design cost us $2500 to build. With a CNC mini lathe that number was $1800. To get there we solved dozens of small design errors, coolant leaking problems, spoke with many different suppliers for quality ball screws, redesigned the control cabinet a few times for both functioning and easy replication. Mini lathe turned CNC as a test. The control cabinet has been cannibalized for a new 1200lb lathe with a pneumatic chuck and live tooling which we're building right now.

Currently we're working on fully new designs that will solve all problems we encountered and will use more powerful spindle motors. These machines will cost us ~$7000 but cut down machining cycles from 45 min -> 8 min for a 4x2 inch part, give us a much better surface finish, better concentricity, better precision in the critical dimensions, and will be more amenable to 24/7 lights out production of parts. A machine with similar capabilities will easily run you over $80,000.

The capability of these machines means a single machine operator can run 20 machines instead of a stretch of 6, and the capital efficiency means we can run dozens of machining lines for the price of one. Machined parts are by far the most expensive part of the BOM and is the main bottleneck to increasing throughput. This change alone will give us pricing headroom for a long time.

I don't want to underplay the challenge here--this took us a really long time to get right. Our cofounder and CTO Dan who comes from a line of machinists and tool makers has been building machines for almost 2 decades and his skill and experience is what made this possible.

Assembly

Motor assembly is winding the stator, press fitting subassemblies, magnet insertion, and final assembly and attaching retaining C clip. I used the stator winding and magnet insertion machines I'd already purchased for this batch, but we spent a considerable amount of time making our own in-house versions. You can see an early attempt that smashed the magnet as it pressed it in (pneumatics can be tricky!) An almost perfect magnet insertion cycle.

At this point we have a V1 of the magnet insertion machine that works and can reliably dispense glue and push arc magnets into the rotor assembly in 60 seconds, built for $900. The machine I'd purchased cost me more than $12,000 before shipping and took 45 days from the date of purchase to be ready to be shipped. This is not even including the extra time needed to have our rotors physically made and sent to their factory. The winding step is quite straightforward. The machine we purchased for $23,000, once set up properly, can do 24 stators every hour. Conceptually nothing wrong with it, but we're confident we can build one for less than a fifth the cost with a self feeding mechanism around it so that it doesn't require a person to tend to it every 5 minutes.

At this stage 24 stators/hr will not be our bottleneck, but as we increase the number of SKUs offered (different stator sizes, different kv ratings) changing the tooling and fine-tuning the tension on the wires will take a significant amount of labor time. Adjusting the tension on the wires is more art than science, so you'd need someone trained up on these machines and they'd have to do this work for a long time before they become good at it.

Just building a new winding machine, which we're confident we can do for less than a sixth the price, will let us offer motors with different kv ratings a lot easier. The machine we purchased also took over 80 days between placing the deposit and it being ready to ship, making it in house will let us stand up new production lines within weeks not months. Sidestepping this issue alone saves millions at even moderate scales.

I want to reiterate this point for emphasis: costs related to machine unavailability or downtime add up quickly, and they can become crippling. Problems like these are very predictable and very hard to mitigate, and thinking you can get away with just good planning, is bad planning.

Testing

This is an area you don't want to DIY things. If we are to remain honest all our metrology and QC needs to be the industry standard, at least in the beginning. The tooling we intend to build around this is just test automation. We're building a system where each subcomponent and subassembly goes through individual testing before it reaches the end of the line, then you secure the assembled motor onto a bench, clamp the wiring to a test computer, and hit go. I'll share more on this in later updates.

Front Office Work

This is a loaded term, but I'm primarily talking about factory management, ERP like software. Essentially, at scale we want to be able to track all inventory, all vendors, all instances of raw material purchases, all machines (purchased or built in-house), tools, products, and production processes. We want to be able to answer questions like: “How many of product A can I make with the raw material that's stored in Factory X?”, or “For the product with tracking ID xxx-yyy, who was the ball bearing supplier?” There is a lot more that good software can do for the production process and we're slowly building it along with the rest of our machinery. We call it HQ. Within it, we want to be able to do:

  1. Everything a traditional ERP does for tracking inventory and predictive analytics.

  2. Full 3D models and assembly instructions for all machines we make in-house. This will include how each part is made and put together.

  3. Full 3D models and assembly instructions for all the products we make. It's quite straightforward right now to put together motors and actuators, but we will be making products with more complex assemblies. This feature will really shine then.

  4. Every time there is a breakdown in a machine (built or purchased) we will document all its details and how we fixed it, with a natural language interface built on top to identify and resolve similar issues going forward. We will also maintain a per-machine history so we can proactively conduct inspections or build redundancy so production lines never fully stop. Telemetry built into our machines will feed data into HQ that'll let us track throughput, detect failures, and later auto-fix issues from a central control panel. This becomes more useful as we build later versions of our machines that come with articulated arms that can be remote-controlled.

  5. A full fledged Resource Constrained Project Scheduler. We clearly define each production step, what machine/tool it takes, how long it takes, how much active attention it requires from the operator and what skills they must have, the context-switching cost of going from one process to another, intermediate yield losses, and error bars around each step.

Once each necessary step is defined, the directed acyclic graph for the full production process for a single product/SKU built, you can run Monte Carlo simulations to know how to scale production.

We will be able to very clearly answer questions like “Is every factory worker properly equipped with all the machines they need?”, or “What set of machines and how many people will we need to make X thousand units of a product from factory A?”. Advances in LLM-enabled natural language interfaces, AI coding, and passive multi-modal analysis a background agent can do for the diligent data entry work that'll be needed to make this system work, have all made this work a lot easier than ever before.

This work isn't incidental to building hardware, it will be essential to our ability to scale production. Having the HQ system act as a central nervous system that keeps all the production “tribal knowledge” organized and accessible will keep production lines coherent. It will prevent drift in quality between different production lines or between different factory locations. Knowledge of how each machine is made, how it works, how it can go wrong, and how to fix it when it goes wrong, will be critical to rapid scaling.

This knowledge collection and transfer process will need to be seamless so that even during times of high stress and time crunches, important information isn't lost to the aether. Not to spam the AI button but AI will actually help a lot in this regard. The HQ system will be built like any other high quality software product, and the features needed and the interface design will be built with tight feedback from the factory floor. This is a long term investment that we also started building in the last few months.

Last Word

If you spend even a little amount of time looking at how anything gets made at scale, and how things get cheaper, it will become clear to you that this isn't just a better way, but that in a long enough horizon it is the only way to mass manufacture. Looking at ever increasing cost of capital, inflation, decrepit state of machine tool manufacturers and their sclerotic sales process, and ruthless competition from China, to proceed as is would be a denial of reality. The only way to do this is doing it the hardest way possible. If this is a challenge you are ready to take on, join us.