Goods-to-Robot Picking: What’s the Catch?

Almost every time I talk to customers about large goods-to-person systems, the conversation eventually turns to robotic picking. So I thought I should summarize some of the key challenges of integrating robots into picking systems. And this time, I made a video out of it as well. You can find the transcript of the video below. Enjoy!

Goods-to-Robot Picking: What’s the Catch?

The general idea of goods-to-robot systems is not bad: you take a goods-to-person system and you replace manual pickers with robots to further reduce the amount of labor hours the system requires. Given the labor shortage in many places, that seems like a good idea. In this video, I explain why it’s not quite that simple and why robotic picking is still a long way from becoming a common technology in warehouse automation.

What is a Goods-to-Robot System?

If you watch this video, you probably know what a good-to-person system is. In a goods-to-person system, an automated storage and retrieval system supplies storage totes to the picking stations, and pickers standing at the picking stations pick from those storage totes and place the items they picked in order totes or cartons. Both the order totes and the storage totes come to the picking station and the picker doesn’t have to move to them. Since we take walking out of the process, picking becomes very efficient. One picker can easily pick between 300 and 600 order lines; of course, the exact performance always depends on the circumstances, such as the type of goods to be picked, the order structure and the design of the picking station. 

And in a goods-to-robot system, you replace the manual picker with a robot. That’s the basic idea.

Piece picking robots don’t cost a fortune. For about €100.000 or so you can get a state-of-the-art piece picking robot. The problem, however, is that you need to have some more things in place to use this robot. And the real challenge of robotic picking is the integration, not the technology.

Integration Challenges of Picking Robots

First of all, you need an AS/RS. And that’s expensive. And that will always be the case. If you’re going to use picking robots in a goods-to-robot setup, you need a powerful AS/RS to feed them. 

With the pick rates touted by some vendors (and of course we can ask how realistic they are, but that’s another story), you could dedicate an entire aisle of single-level captive shuttles to each robot. So that’s about €1 million plus per aisle. Add to that another €100,000 to connect the robot to the rest of the conveyor loop.

There is more cost. But before we can talk about that, we need to talk about another challenge first. The cost of using robots and how useful the robots can be depends largely on the order structure.

In case you aren’t familiar with the term order structure, I recommend you watch the introductory video of my video course on order structure. In a nutshell, order structure refers to the composition of orders in terms of the number of pieces per orderline and the number of orderlines per order.

And this is important because the more orderlines we have per order, the more likely it’s that among all the orderlines there is a SKU that the robot cannot pick because it’s outside its vision or grasping capabilities.

Now we have two choices: Either we send every order that contains orderlines the robot cannot pick to manually operated picking stations. This is the simplest option, but it also has the disadvantage that the robot might end up having little to nothing to do if most or all of the orders have many orderlines and there is always an orderline among them that the robot cannot handle. 

The second option is that the robots pick whatever the robots can pick, and humans pick the rest. But then it’s necessary that these orders are consolidated.

For the consolidation, again, we have two options. The first option is that we pass unfinished orders from the robotic stations to the human-operated picking stations where we finish them up. Quite frankly, this is a control nightmare and introduces significant buffering and sequencing requirements and unless you know exactly what you’re doing, I do not recommend you try this out. 

The reason this is complicated is that we now have to synchronize unique (i.e., non-virgin) order totes with storage totes at the human-operated picking stations. It also can go wrong in terms of capacity planning for the sequencing and the buffering because we do not know in advance how many orders we’ll need to route this way. 

The second option is that we pick orderlines at robotic and human-operated picking stations independently and simultaneously and send them to dedicated consolidation stations, and these consolidation stations could also be our packing stations. This makes the planning of the packing stations a bit more complicated and the stations much more expensive, but from a control point of view it’s simpler.

And by the way, you can already see that you are certainly not going to have pick & pack workstations when you introduce robotic picking, but instead you will have to separate picking from packing. 

So, in the case of multiple orderlines per order, you need consolidation stations (€) or extensive conveyors for buffering and sequencing (€€), and in both cases, you need to make costly upgrades to your warehouse management system and warehouse control system (€€€).

Long story short: stay away from goods-to-robot picking if you have many orderlines per order and often have at least one SKU per order that cannot be picked by the robot.

Oh, and there’s another cost: the cost of teaching the robot dealing with new products. Of course, you can’t assume that you install a robot and it just picks. It has to be taught how to pick, and in the worst case that can be different for each individual product. 

Some robot suppliers claim that their robots are self-learning. They’re able to learn unsupervised. They teach themselves how to pick and place products in … their spare time. There are two problems with this:

  1. Why does your robot have spare time? It probably shouldn’t.
  2. There are certain characteristics of products that a robot cannot identify on its own.

First, why shouldn’t the robot have spare time? If you want the robotic picking system to make financial sense, you want to run it at least two shifts a day, maybe even three shifts. In any case, the robot should have no free time during normal operation. It should not be idle. But let’s say you normally run the system two shifts a day, and you want to let the robot use the third shift for learning. 

Well, you can certainly do that. But if something goes wrong, you want someone to be there to fix it. Otherwise, you run the risk that (a) the robot will just stop and not learn anything, and (b) the robot won’t be available for its normal work shift the next morning.

The second argument, however, is even more important. Because if the robot is given time for unsupervised learning, it’ll never be able to get the full picture. The robot can take photos of the new products. It can grasp the new items from different sides and angles and put them down again.

However, it cannot automatically know if, for example

  • there are certain sides from which it should NOT grab a product.
  • It cannot know how sensitive a product is, but that is important for placing the product in the bin.
  • In the same way, it can’t know if it’s allowed to put another product on top of that product.

So I’d say that there is always a certain amount of human interaction involved in the teach-in process. It can be quick if the teach-in process is well designed, but it’s not going to be free and therefore it’s always going to have a cost.

Conclusion

Picking robots in a goods-to-robot setup can provide an economic advantage by improving productivity. After all, they replace labor, thereby lowering your operating costs. The economic benefit is very likely much less than you’d expect. The capital expenditure is much higher than you’d expect and the savings in terms of labor hours are lower. The real challenge of robotic picking is the integration of the robots in the system, not their technology.

And using robots in goods-to-robot picking makes more sense under some circumstances than others.

What are the Right Circumstances for Goods-to-Robot Picking?

Goods-to-Robot Picking picking will be most useful if

  1. you’re planning on deploying a huge AS/RS for goods-to-person picking anyway
  2. the robots can cover a base demand
  3. there are always staff available for trouble-shooting at the robot anyway
  4. there are always staff available to pick all those items which the robot cannot pick
  5. most orders have very few orderlines, preferably only one, so that no consolidation of robot-picked orderlines and human operator-picked orderlines is necessary
  6. you already separate pick and pack
  7. products are insensitive and can be dropped for faster picking rate
  8. the product portfolio hardly changes at all
  9. all products are in ambient temperature range and are not moist

If you can tick all of these boxes, you should think about goods-to-robot picking. If any of these boxes remains unticked, you should think.. at least twice and you should be very critical about the promises the robot sales guy makes. 

Fixing the Non-Broken Part 

You spend 20 million Euros or more on a shuttle system or an AutoStore. 

And then you want to replace a guy 

  • who costs you 35k per year,
  • who on average picks faster than the robot, 
  • who can pick any SKU,
  • who does not have to be taught how to pick a new SKU, 
  • who can pick and pack in one go, so you don’t have to split those processes
  • who can be used in almost any process in the warehouse because he’s pretty flexible and versatile, which makes staffing much easier,
  • who does not need cold redundancy or a fallback solution in case he breaks or needs maintenance, 
  • and you spare yourself all the mess with order consolidation.

All in all, with goods-to-robot systems it seems like you are probably trying to fix the non-broken part. And that’s my main criticism of the concept.