By Alberto Moel, Vice President Strategy and Partnerships, Veo Robotics

Welcome back, fearless reader, to the third installment of our series on the economics of automation. Our previous postings first made the case that full-on automation is inflexible and fragile, and then made the point that high-levels of automation can be terribly uneconomic. We also discussed the fact that an all-human manufacturing approach is equally suboptimal under many reasonable conditions—the best outcome is a mix of humans and machines working together.

In today’s post, we’ll go into detail on this topic with a realistic example in auto parts manufacture. We’ll be in the weeds for a while, so if you’d rather just skip to the end and look at the final chart in this post, you will see that, for a wide range of production volumes, human-machine collaboration leads to the lowest costs in a manufacturing process relative to either all-human or fully-automated assembly.

Which brings me to a small but important detour on the definition of optimality in a manufacturing process. Manufacturing, broadly, is the process by which physical goods get transformed into other physical goods. Besides the cost of those input goods (say, the raw materials), there is the cost of the manufacturing process itself. This includes not just the actual direct labor, but also the tools and equipment required for the transformation; overheads such as supervisory labor or electricity; the parts, fixtures, and consumables required; and the engineering and design effort needed to put together, test, and upgrade the manufacturing process. For simplicity, we’ll define a manufacturing process as optimal if the sum total of the manufacturing costs per unit of production are at a minimum.¹

How to estimate each of these costs requires us to put on our green accounting eyeshades—the standard techniques aren’t anything revolutionary but merit some simple explanation. Direct costs, such as direct labor or consumables, are estimated by figuring out labor rates and such per hour and dividing by the number of units produced per hour. Simple enough.

The costs of equipment are allocated via depreciation, which is basically the cost of the equipment divided by the “useful life” of the equipment relative to what’s being produced. This is a bit tricky, as “useful life” depends on whether the equipment can be repurposed in different manufacturing steps, such as, for example, whether a robot used in a manufacturing workcell for automobile assembly can be reused somewhere else once that process has been finished.

In the case of a robot that costs $50,000 and has a useful life of maybe 100,000 hours,² each hour of production “costs” about 50 cents of the robot’s lifetime. And if you’re making 100 widgets an hour (not an unreasonable number), that’s half a cent per widget. On the other hand, if you’ve designed and built a set of fixtures to hold parts for the robot to work on and those fixtures cannot be repurposed in another manufacturing process, their useful life is just the length of the production run. Hence, if you spend $10,000 in fixtures, but only make a total of 10,000 widgets with them, the manufacturing cost attributable to the fixtures is now $1.00 per unit.

And lastly, we have the costs of the engineering and design effort to put together, test, and upgrade the manufacturing process. This is basically a non-recurring one-off expense³ that is amortized⁴ over the production run. It can sometimes be multiples of the actual hardware costs and it drives much of the economics of manufacturing, as we shall see.

OK, enough accounting nerdery and perambulations. What’s this realistic example of human + machine prowess I keep talking about? Very simply, it’s the insertion of three rubber bushings into a machined suspension knuckle. What in the wide world of sports is a suspension knuckle? Well, it’s the part that holds your car’s suspension in place and, for some unfathomable reason, it requires the insertion of three bushings (Figure 1).

The assembly of a capital good such as a car, washing machine, or jet engine requires hundreds or even thousands of these mindless but highly-specific and necessary steps. Sometimes it’s three bushings in a suspension knuckle, sometimes it’s two screws holding a cover together, and other times it’s a bunch of wiring harnesses that need to be threaded inside a metal fixture.

Simple enough, right? Not so fast. Because the rubber bushings are pliable and soft, they are hard for mechanical tools to grasp, position, install, and set into the knuckle easily. It can be done, but it’s a tough ask for a machine and probably won’t have very good reliability. Which is why, in general, steps such as these are performed by humans.

An additional problem is that that this simple step must be accompanied by all kinds of side dishes and interesting little dances, such as those implied in Figure 2. In order for a human to mount the bushings in the knuckle, the knuckles have to be available and the human has to do all kinds of motions besides inserting the bushings: walk to the bins with the knuckles, pick up the heavy (30-pound or more) knuckle, and mount it in the fixture. Then, after inserting the bushings, take the finished knuckle (again, 30+ pounds) to the output bin. Lather, rinse, repeat 90 times an hour, 8 hours a day. Great fun!

We estimate that this process cycle takes about 45 seconds, of which 25 involve picking stuff up, putting stuff down, or walking around while carrying stuff—effectively more than half of the direct labor is materials handling. While it’s certainly flexible—if the part design changes, the human doing the work can adapt on the fly and there’s no need to re-configure anything, except perhaps the simple mounting fixture they’re using—in the end, it’s clear that this manufacturing step is a poor use of human labor. Too much human effort is wasted, not to mention the risks of repetitive stress injuries, accidents, or just plain boredom.

So, let’s say you decide to take the human out of the loop completely in this process—humans are expensive, unreliable, get bored easily, you name it. And let’s assume you are able to develop robotic and automation technology that can reliably grasp the rubber bushings and mount them just right into the knuckle.⁵ What would your workcell look like? Probably something like Figure 3.

Note the additional rigamarole to replicate the loading and unloading a human can do so easily, the extra fixturing so that the robot can reliably mount the bushings into the knuckle, and the systems and quality control hardware to make sure the robot does a good job.⁶ Sounds like a lot more work to get this workcell to work.

But there are some clear benefits to fully automating the workcell: the robot can work much faster (we estimate a cycle time of 20 seconds) and it never gets tired, meaning there are labor cost savings. If done correctly, there are likely some quality benefits, too (perhaps higher yields and lower scrap rates). The question is whether these benefits are enough to overcome the incremental fixturing, programming, and design work of the more complex workcell.

By now, it should be clear to you that humans are good at some things (dexterity, judgment, flexibility) and machines are good at others (precision, reliability, tirelessness, speed, strength). So why not mix and match, bringing the best of the human and machine together? In our simple example, the obvious arrangement would be for the robot to bring the parts in and out of the bins while the human mounts the bushings, as in Figure 4.

In this setup, the robot loads and unloads in parallel with the human installing the bushings in a dual-fixture setup. This means that the human only needs to walk between the two fixtures while the robot quickly brings parts in and out of the work area. The amount of rigamarole and fixturing is reduced here compared to the fully automated workcell as it’s the human (with all of their skill and dexterity) that’s mounting the bushings, not the robot.⁷ We estimate that the cycle time of this arrangement is about 25 seconds, somewhere in between the fully manual setup (45 seconds) and the fully automatic one (20 seconds).

So, of these three ways to complete the “simple” process of installing a rubber bushing in a suspension knuckle, which is optimal from an efficiency perspective? In other words, which one has the lowest manufacturing costs per knuckle unit? All three have tooling and hardware costs—either a simple fixture or a fully-automated robotic handling system—that get depreciated on a per-unit basis. All three have labor costs—even the fully-automatic version has some overhead labor costs associated with equipment oversight and maintenance—that get allocated on a per-unit basis. And all three have design and workcell development costs—from the simple “get me a couple of bins and a table” to the complex robotic workcell development effort—that get amortized per unit.

But the three manufacturing approaches also have different efficiencies, as represented by varying cycle times or scrap rates, so the tradeoff is between fixed and variable costs in workcell economics. We can capture the basic (but realistic) inputs and outputs of our model in Figure 5, for a total production run of 10,000 units.⁸

What we see here is that for a 10,000-unit production run, the lowest cost per unit is with the fully manual process. Even though the fully-automated process has much lower labor costs per unit ($0.08/unit as opposed to $0.78 for the fully-manual process), its much higher efficiency doesn’t overcome the substantial hardware and non-recurring expenses (NREs) needed to automate the process. The mixed human + robot process we advocate is also a bit inferior—although the hardware and NRE costs are lower, they aren’t low enough to be preferred over a fully-manual operation. But what if we had a much longer production run, such that all these design and hardware costs could be better amortized over many more units of production? Figure 6 shows the condensed summary results for a 100,000-unit production run.

In this case, the fully-automated process wins. Although the total hardware and NRE costs per unit are still higher than the fully-human or human + robot processes, the higher efficiency and lower labor costs are enough to overcome these fixed costs.⁹ Note also that for this high production volume, the mixed human + robot workcell design is also a better bet than the fully-human process.

As promised, there is a wide range of production volumes where the human + robot workcell design is better than either alternative! As we vary the production volume, we see that for a range of about 25,000 to 70,000 units of production, humans working together with machines is the best approach (Figure 7).

As a matter of fact, if we have no idea what our eventual production run will be, the lowest average expected cost per unit can be reached by designing the workcell to be collaborative between humans and robots.¹⁰ In other words, humans working together with machines beat either working alone, and if you don’t know how many units you’ll be making, hedging your bets by building collaborative workcells is the best approach.

¹ This may sound like pedantry, but there are many other ways to determine whether a manufacturing process is optimal, including lowest cycle time, highest return on investment, highest yield per unit time, and many others. All of these are related and, in most cases of practical interest, the lowest manufacturing cost per unit of production is likely to be near or at the lowest cycle time, the highest ROI, or the highest yield per unit time.

² I’m not making this number up: 100,000 hours is 11.4 years of 24/7 operation, which is about right for a high-quality robot. Try driving your car non-stop for 11.4 years. See how far you get. Robots are amazing pieces of machinery.

³ Known in manufacturing circles as a Non-Recurring Expense (NRE). Go figure.

⁴ Amortization is just a fancy word for depreciation when what’s being depreciated isn’t an actual physical good, but an intangible, such as an NRE.

⁵ This, friends, is far from a fait accompli. We’re still in the baby steps phase of technology that can reliably grasp soft, pliable materials and insert them in an orifice as well as an ambulatory two-year-old picking up a dirty candy wrapper and sticking it in his mouth.

⁶ And I am being generous here. In reality, such a workcell would probably require two robots, as the loading and unloading end effector is not very good for bushing mounting and vice-versa. Having the robot switch end effectors every cycle is extremely inefficient.

⁷ To be clear, this kind of setup is unusual in actual manufacturing workcells. The idea of having a robot and a human share the same space is relatively new and standards and practices are still evolving. One option is using a collaborative robot to safely interact with humans, but this would greatly limit the range of automated actions to lightweight payloads and slow speeds. Another option is the forthcoming Veo Robotics system (obligatory plug) to make traditional industrial robots collaborative.

⁸ In other words, assuming the useful lives of the workcell-specific fixtures end when 10,000 units have been assembled. Some of the hardware, such as the robot, can be reused elsewhere, so the robot’s useful life can be modeled as the actual useful life of a robot: 100,000 hours (give or take).

⁹ To be fair, the estimates for workcell fixturing costs in the fully-automated case are probably conservative. Increasing these fixturing costs to, say, $60,000 means that the breakeven where the fully-automated process is superior to the human+robot process is now over 120,000 units of production.

¹⁰ I’ll spare you the math, but it involves assuming a distribution of production volumes and figuring out the mean cost per unit. If you’re certain that the production volume will be low, you would not automate the cell. Conversely, if you were certain of a long production run without any product changes, automating it completely is best idea. But anything in between, or a high level of uncertainty on how many units to make, calls for human-robot collaboration. And you can take that to the bank.

Human-Robot Collaboration Case Study

© Copyright 2022 Veo Robotics, Inc.