Author: Nikolaus Correll

Learning “lean” starts at home

Learning “lean” starts at home

Some people say lean manufacturing cannot be learned, it can only be experienced. I agree with this statement, as “lean” is not just a collection of rules. Worse, more often than not, the opposite is true and what has been the “golden rule” nine out of ten times, the tenth time it will be your worst enemy. This is captured well in Christoph Roser’s article “What is the true North in lean?” who uses the analogy of a compass needle that gives you very good directions to the North pole until you get actually close. Then, the compass needle will point you to the magnetic north pole (which is not quite where the geographic pole is) and eventually let you turn in circles.

This does not mean that you cannot experience “lean” by immersing into a good narrative (not what you get today) or by putting theory into practice at home. One of my least favorite tasks is emptying the dishwasher. I cannot really describe why it is so much worse than loading it. Maybe it is because I can feel that I somehow waste my time, but cannot say how.

This article will use the dishwashing task to explain some of the key tenets of lean thinking by

  • Letting you ask the right questions. What is your goal?
  • Introduce the “seven deadly wastes”, strong indicators of things going not lean.
  • Illustrating why “lean” is a continuous improvement process and why simply following rules will not work.
What is your goal?
Clean dishes, the symbol of “work done”, often get in the way of dirty ones.

What it is that makes unloading the dishwasher so annoying? Answering the question on what to optimize is never easy. In manufacturing, you want to manufacture things as cheaply as possible. Do you really? Or, do you want to make things as fast as possible? Indeed, whether you need to optimize for efficiency, i.e. the ratio of what goes in and what comes out, or throughput, i.e. the amount of goods that come out, really depends on your business.

As we cannot really pinpoint what it is that makes unloading the dishwasher so unrewarding, let’s focus on getting it done as quickly as possible. Time is easy to measure using a stopwatch. How can we go about decreasing time then? Is it to try to move as quickly as possible, risking that dishes break every now and then, leading to defects, this first of the “seven deadly wastes”? That might be not sustainable. We are rather interested in simple changes that do not lead to increased tear and wear on both dishes and people, but being able to do more with less.

The first exercise is to identify the different subtasks that emptying the dishwasher entails. The first is taking items out. This is where we really make progress toward our goal. The second is placing items in cupboards and shelves. Finally, we need to move items between the dishwasher and their storage location. Transport is always waste in lean thinking. It is quite easy to detect and reducing it usually goes a long way.

This leads to the first (quite obvious) insight to store dishes as close to the dishwasher as possible.

For example, left or right above the dishwasher for light dishes and left or right next to the dishwasher for heavy pots minimizes not only transport, but also motion. Motion is waste, and optimally one would elevate the dishwasher just enough that bending down is reduced. All of this is often not possible in most kitchens, and most adventures into “lean” often end when the obvious improvements are not applicable. Fortunately, the dishwasher problem – as most trivial looking problems in manufacturing – is surprisingly deep.

Giving up now is actually fatal. Not having access to an optimal storage location is already a problem. Even more effort should therefore be put into optimizing your second choices. Most people don’t do this, however, as they underestimate the big impact that the sum of small changes can have in the long run.

Reducing the number of parts, tools, and product variations can go a long way in manufacturing or make for a “lean” camping trip. (C) REI.

The first question you should ask yourself is whether you can get away with a more compact assortment of dishes and cutlery that can indeed be stored in a more optimal location. All sorts of excess inventory is usually waste, which includes “work-in-progress”, such as dishes sitting in the dishwasher waiting to be loaded. While the answer here is “no” for most people, and there indeed is an occasional need for even the fanciest sauciere, this leads to a consideration of frequency of use.

Frequently used items should be stored in the closest location, those less often used should be stored in the furthest available location.

This is probably already the case for most people and if it is not, this has probably been one of the most useful articles you have read in a year. But there is more we can do.

Now that we have minimized the overall motion that is required to solve the problem by optimally assigning item groups to a set of locations (a computationally hard optimization problem known as the “quadratic assignment” problem), it is worthwhile to take a closer look at what actually happens during that wasteful motion.

Our first realization is that we would want to carry as much as possible everytime we go. This becomes pretty clear when imaging that walking back and forth between the drawer and the dishwasher for every single spoon will take much longer than moving the cutlery by the handful. This directly informs how we carry out the emptying task – we should try to grab as much items as we can before going anywhere.

Many more items can be carried when stacking them, we should therefore focussing on unloading items in groups of identical objects.

It turns out that some items stack better than others. For example,

The “Masskrug” apparently stacks very well, even when full. The world record at the Oktoberfest has been waiting 27 at once.

I can easily grab 10-15 spoons in a single hand or stack 10 or so dishes, but I have a much harder time with glasses, moving only two in each hand. Glasses also don’t stack well with mugs, which don’t fit with wine glasses and so on. This introduces a second “frequency” in our consideration. Some groups of items, such as dishes, can be moved during one or two trips, whereas glasses and mugs require many trips.

Groups of items that require many trips from dishwasher to storage location should be closer than groups of items that can be stored in fewer trips.

(Readers already familiar with lean will notice that batch sizes of one are usually a “golden rule”, but are always a trade-off with motion and therefore not desirable here. This is a typical conundrum in lean thinking, where blindly applying a rule would be very bad.)

Now that we better understand how to speed up the unloading

Inside of a dishwasher stacked with clean plates, bowls and mugs at the end of a washing cycle in a health and hygiene concept

process, we might also want to think about how to load the dishwasher. In an ideal scenario with storage to the left and the right of the machine, we should load items accordingly. As we want to grab as many items as possible and stack them, items should also be loaded this way. This can usually be achieved without any additional cost, but pays off downstream.

Has everything been said about the dishwasher unloading problem? Far from it.

“Lean” is a continuous improvement process. This is because some improvements will also only become obvious once the first steps are done.

For example, now that you have minimized motion, it might make sense to think about moving item groups on a tray. This bears the question whether reduced round trip time justifies the additional manipulation (placing and picking from the tray). This problem can probably only be solved using a stopwatch. It is also interesting whether motion could be completely reduced by working with a partner who is strategically placed and to whom you can hand over items. Whether this approach is more than double as fast as when you do it alone (super-linear!) will depend on whether you can balance your line, that is reduce waiting  – another form of waste – for either you and your partner while one is storing or retrieving an item. Also, now that every second counts, opening and closing cupboards and drawers becomes a nuisance. While good organization and planning can help to open and close each cupboard only exactly once, some might start thinking about structural changes and remove unnecessary doors.

Lean makes it easy to get carried away in details, however. There are two additional wastes that are easily overlooked in a smoothly running system that has been tuned by a genius. First, not everything in the dishwasher might actually need washing, but instead could have been stored right away or after a quick wipe-off. This kind of waste is known as over-processing and is very common, but much harder to detect. But what is really the dumbest thing one could do? Putting dishes away right before setting the table. This is as good as having to throw away what you just made and is known as overproduction. Therefore, always grab dishes from the clean dishwasher and only start it once you start running out on critical items. (Manufacturers manage the equivalent of this process using a so-called Kanban card.)

If you make a living of dishwashing (kind of like manufacturers do by making things), you might also think about structural changes to your operation, creating an optimal setup for your operation. Don’t forget that the dishwasher needs to stay next to the sink, not just for the plumbing, but also for the rinsing, otherwise all your new savings are gone quickly down the drain.


The goal of “lean” is to

  1. Make a process as fast as possible.
  2. Make a process as cheap as possible.
  3. Maximize efficiency.
  4. Maximize throughput.
  5. Whatever maximizes your bottom line.

In lean thinking, what is not considered “waste”

  • Items that break during the process.
  • Unnecessary motions.
  • Time an employee waits.
  • Overprocessing of items.
  • Efficient use of a worker’s time.

Replacing the human worker with a robot that does the job half as fast, but at half the operational cost is a different approach to make this process “lean”. Which statement is correct?

  1. True. Increasing the bottom line is the overall goal.
  2. False. The process in itself is sub-optimal and needs to be improved first.
  3. False. Dishes need to be available as quickly as possible and the lower speed of the robot disrupts “flow”.
  4. True. Replacing human with machine labor reduces uncertainty and makes the process more predictable.
  5. All of the above.

How lot sizes change your efficiency

The saying goes that single-item flow is a lot more efficient than batch processing. This is easy to see when considering that seeing large inventories piling up anywhere in your factory and other stations waiting to process a batch are one of the Seven Deadly Wastes. But how bad is it? Would reducing batch sizes make your process 20% more efficient? Or double? Will it compensate the additional motion and transportation – other deadly wastes or is the cure worse than the disease?

Let’s consider a simple scenario comparing lot sizes of four that have to be processed by three stations. For simplicity, we assume that each item has to be processed individually, which takes one time step, and moving one or all items to the next station also takes one time step. The animation below shows processing the items in lots of four.

Processing items in lots of four at three different stations. Moving the material through the system takes 15 time steps.

We can now compare this with a process in which each processed item is immediately moved to the next station. The advantage is here that all three stations will be able to work in parallel. The difference is dramatic:

Moving items to the next station immediately allows processing the same amount of material in 9 time steps.

Processing the same amount of materials is almost double as fast. This is no surprise, as work is done in parallel, requiring up to three workers, whereas processing the entire lot can be done by a single worker. (The cost of labor remains the same, as the same amount of work overall is being done.)

The key difference, however, is that materials have to be moved 12 times vs. 3 times. This works on a so-called flow line, where stations are next to each other and workers can pass parts from one to the other, but will increase cost substantially in a job shop setup, where each transfer of material requires significant transportation. This is why flow-lines are so much more efficient than job shops.

Would it be worth to put in a dedicated material handler to simulate a flow shop in such a scenario? Let’s do the math real quick. Assume we have lot size L with L=4 in this example. Assume we have S stations with S=3 in this example. Processing the entire lot therefore takes O(LS) time steps with O(S) moves. (Indeed, LS=12 plus S=3 is 15.) The “Big-O” notation is used to analyze algorithms in computer science and reads “on the order off”. It allows us to ignore details like differences in processing time across different stations, but focus on the fact that lot size and number of stations are multiplied.

The time it takes to process lots of size L through S stations is proportional to the product of L and S.

Let’s look a single item flow line now. Once the first item has been processed, it is moving to the next station. The first station is done after L time steps. By this time, the other stations have begun working, however, and the first item is leaving the line after being processed at S stations. The total time is therefore O(L+S). Another way to think about is that the first item will be done after S time steps and the last L time steps later. The total number of transportation events is now O(LS).

The time it takes process L items through S stations using a single-item flow is proportional to the sum of L and S.

Assuming you need to process L=1500 parts through S=6 stations, the difference in processing time is one order of magnitude, or it is almost 10 times slower to process items in a lot. If material handling would be free in terms of cost and the extra head-aches it creates, moving your process time from O(LS) to O(L+S), would be a no-brainer. It would allow you to substantially decrease lead time, rent, and facility cost.

Transportation is never free, however. Workers need not only to move materials from station to station, they would also need to know where to move each item, something this would change with every product being made. A trade-off is to chose intermediate lot sizes. For example, instead of processing 1500 items at the time, workers could work in batches of 100 and then pass on these smaller batches to the next station. Note, that the same math applies: L is now 15 (15*100=1500) and O(L+S) is still better than O(LS), only the difference is less accentuated.

In practice, the batch size B should be chosen so that processing B items at a specific station takes as much time as moving these parts from that station to the next. For example, when using a dedicated material handling worker who picks up processed items every 10 minutes, B should be the number of items a worker can process in 10 minutes.

An elegant solution to this problem (and one that we can help with) is using material handling robots that connect the different stations at a constant rate of flow, picking up whatever number of items already have been processed and moving it to the next. Operating such robots is substantially cheaper than the cost of a worker (up to 8 times in industries such as aerospace engineering) and will automate a lot of the book-keeping, such as when to move materials where.

Reducing lot sizes becomes problematic, when processing times at different stations widely vary. There are two extremes: in one, all materials will need to be processed at once, for example when washing or coating the parts. In the other, processing takes much longer, leading to inventory build-up at this station. The solutions are simple, however. In the first case, the simulated robotic flow line needs to end just before the batch processing station, with another one starting right afterwards (if necessary) to maintain all the benefits. In the second case, additional resources (workers) should be added to balance the line. Here, the robots can help with arbitrating the load, for example, by automatically switching between delivery to two different stations.

Robotic Materials Inc. mobile manipulation solution is available for rent or purchase, contact us today!

Nikolaus Correll is the CEO of Robotic Materials Inc. and a Professor of Computer Science at the University of Colorado at Boulder. He has been designing and building large-scale distributed robotic systems from swarms of robot as small as a ping-pong ball to teams of mobile manipulating robots with fine manipulation skills. He is manufacturing robots for manufacturers in Colorado since 2016.

How to turn your job shop into a flow shop with material handling robots

A job shop is usually the least productive way to organize your manufacturing effort, see for instance “Why Are Job Shops Always Such a Chaotic Mess” by Christoph Roser. Yet, there are reasons job shops do not go away, and this is why you are probably reading this article.

Three products running through different stations of a job shop (left). Three products running through different stations of a flow shop (right). The representations are not equivalent and not every job shop can be turned into a flow shop.

Lets look at the reasons everyone tells you to better turn your job shop into a flow shop:

  • Flow shops are easy to measure and optimize. You can immediately see what goes in, what comes out, how long it takes and where things get stuck.
  • Flow shops are easy to automate. Each step is consistent and repetitive.

In fact, these advantages are so strong, that the advice is to essentially do whatever it takes to squeeze the processes into your job shop into a flow shop where some products simply skip certain stations.

So why would you want to stick to your job shop layout?

  • Your existing processes simply cannot be mapped to a flow shop without adding many redundant stations.
  • Flow shops take more time to setup, in particular if you have to account for all the possible products you are making.
  • Flow shops are not flexible. This becomes a problem when your products constantly change such as for contract manufacturers.
  • Flow shops do not scale well with increasing demand, but are optimized for a certain throughput.

In practice, most manufacturers already employ a hybrid model, setting up multiple flow shops within a job shop. While this provides them with an optimal trade-off in terms of efficiency, it is difficult to organize (workers need to know what to bring where) and measure and optimize overall performance.

There is a way to reap to truly reap the benefits of both approaches. Autonomous material handling robots allow you to turn your job shop into a flow shop, with the robot quite literally operating as a flexible conveyor belt that connect all of your stations equally. For example, a robot might provide a station with a shelf with new parts as well as a shelf for outgoing parts upon the press of a button. Once the process is complete, the robot will first move the end product to a new station, then remove the new incoming parts shelf.

Furthermore, the operator can obtain detailed statistics of not only what went in and what came out of a station and when, but also visualize the entire flow of an evolving product throughout the line. This is data that is historically easier to gather from a flow shop, but not necessary available.

Robotic Materials Inc. mobile manipulation solution is available for rent or purchase, contact us today!

Nikolaus Correll is the CEO of Robotic Materials Inc. and a Professor of Computer Science at the University of Colorado at Boulder. He has been designing and building large-scale distributed robotic systems from swarms of robot as small as a ping-pong ball to teams of mobile manipulating robots with fine manipulation skills. He is manufacturing robots for manufacturers in Colorado since 2016.

From Mainframes to PCs: What Robot Startups Can Learn From the Computer Revolution

In their search for killer apps, robotics companies should look at the amazing evolution of computers

By Nikolaus Correll

In their search for killer apps, robotics companies should look at the amazing evolution of computers

Autonomous robots are coming around slowly. We already got autonomous vacuum cleaners, autonomous lawn mowers, toys that bleep and blink, and (maybe) soon autonomous cars. Yet, generation after generation, we keep waiting for the robots that we all know from movies and TV shows. Instead, businesses seem to get farther and farther away from the robots that are able to do a large variety of tasks using general-purpose, human anatomy-inspired hardware.

Keep reading on IEEE Spectrum…

Robots Getting a Grip on General Manipulation

How a new generation of grippers with improved 3D perception and tactile sensing is learning to manipulate a wide variety of objects

Robotic materials gripper
Photo: Robotic Materials Inc.
A gripper created by Robotic Materials Inc., founded by the author, Nikolaus Correll, performs a manipulation task during the industrial assembly competition at the World Robot Summit in Tokyo.

This is a guest post. The views expressed here are solely those of the author and do not represent positions of IEEE Spectrum or the IEEE.

While robots have prepared entire breakfasts since 1961, general manipulation in the real world is arguably an even more complex problem than autonomous driving. It is difficult to pinpoint exactly why, though. Closely watching the 1961 video suggests that a two-finger parallel gripper is good enough for a variety of tasks, and that it is only perception and encoded common sense that prevents a robot from performing such feats in the real world. Indeed, a recent Science article reminded us that even contact-intensive assembly tasks such as assembling a piece of furniture  are well within the realm of current industrial robots. The real problem is that the number of possible manipulation behaviors is very large, and the specific behaviors required to prepare a club sandwich aren’t necessarily the same as those required to assemble a chair.

Keep reading on IEEE Spectrum…

Proximity and force sensor nominated for “Best Paper” Award

Proximity and force sensor nominated for “Best Paper” Award

The paper “Integrated force and distance sensing using elastomer-embedded commodity proximity sensors” by Radhen Patel and Nikolaus Correll was nominated for a “Best Paper” and a “Best Student Paper” award the 2016 Robotics: Science and Systems Conference in Ann Arbor, Michigan. RoboticMaterials has begun negotiating a licensing agreement with the University of Colorado’s tech-transfer office.

Theme: Elation by Kaira.
Cape Town, South Africa