Cross Webinar - Robotic Bin Picking

In today’s manufacturing environment, the availability of good labor is at an all-time low. To address this, automating simple repetitive machine tending is a great place to start, as it frees up existing labor to do higher value tasks. 

Our automation team has tremendous experience working with solutions in a variety of settings and has developed an ideal solution to streamline your process.  In this webinar, we'll discuss 3D robotic bin picking solutions and how they can benefit your operation.



View and Download Webinar Slides

Take a look at the webinar transcript below


Why are we here?
It’s because you may have an application, or you’re just interested in the technology of bin picking so what we are going to talk about today is bin picking and you. This is roughly how it is going to be laid out: first I’ll go over who Cross Robotics is, what is bin picking, what is involved, how the technology provides ROI, which is going to be important because as I’ll say many many times it’s not the cheapest or the easiest way to present your parts to a robot, when you should consider, when not to consider, sweet spots, kind of indicators, things that bin picking kind of jives with well. What we have in place that might be able to help you, we’ve put together kind of a solution with a cobot and a bin picking camera, and then right after we’ll do a live demo and take any questions. 

 

Who is Cross Robotics?
Well obviously, we do robots. We can provide robots, program them, integrate them. We can provide end effector components, design general component, tree safety, area scanners, light curtains, extruded aluminum workstations, just mechanical design, all the cart services and support. We want you to consider us as an extension of your engineering department. So if you have a robotic application or even part of one that you want our help with, we will be as involved as you want us to be. At the end of the day our focus is to help you achieve your business goals by focusing on your most challenging applications. 
 

What is bin picking?
Well a spoiler alert, kind of, pretty much what it sounds like. It’s picking parts out of a bin. In a nutshell you have a camera system, something to control the camera, and a robot. It doesn’t actually need to be like a six-axis robot but it helps because your parts can be oriented but anyway. So you have a really really sophisticated vision system see the contents of the bin and pick the best candidate to pick up. Imagine something like you guys have played those rigged claw machine at the arcade, the process is kind of similar. You kind of scope out what’s in front of you, you can’t predict anything but you can see what’s in part of you, what you’re going to go for then guide an end effector and do some kind of approach and grab process. The only difference is a robot isn’t clearly rigged in every way and it just doesn’t arbitrarily drop part because it was designed to kind of the opposite. Bin picking is really easy for people. In fact, I could blindfold you, put you in front of this bin and without even telling you what the parts are you could probably pick one part out of the bin no problem. It’s actually so easy if I then said hey great job you’re going to do that for eight straight hours and you get an hour for lunch you would probably wince pretty hard at the prospect of doing that. We’ll talk about the ramifications of that a little later. But bottom line is it’s really easy for people, has been really difficult to robustly automate.

 

Part presentation
Anyone in automation, either designing machinery or implementing automation you know that part presentation is a major major part of any automation effort. How are you going to present one part at a time, predictably to a robot, work cell, conveyer belt or camera. So a lot of time, effort, and money probably most importantly goes into this and bin picking is just one of several ways to handle this. 

 

Other ways to handle bin picking
I have listed a couple, nested trays like an egg carton or some kind of jig or fixture, bowl feeding for you know or less fixturing require still pretty fast cycling times, 2D vision as you can see there their parts just laid out on a table. You have your fixed in your orientation you just a camera to tell you X and Y which is pretty straight forward as far as vision goes. And you have flexible feed which some of you guys may be familiar with but basically you have your part on a table but you have the ability to shake the table so if your camera doesn’t find parts you have the ability to agitate them a little bit to find something to create a new lay out for your camera. Again bin picking is one of several ways to handle this. I would encourage most customers to investigate at least these other ways of doing it because they are more straight forward and can be cheaper than bin picking. 

 

Investment considerations
Here are a series of radar graphs and this is what each of them looks like. So the more the radar graph is full to the outside, the closer to the outside it is the “better” it is so initial investment, maxed out means its not too demanding as far as capital expenditure, expertise need. Maxed out means you don’t need that much expertise. So think of it in terms of better or worse in terms of some kind of absolute value right. So if we look at kind of the different ways of doing it um the main thing I want you guys to take away from this is look at the differences between mechanical singulation I’m pointing at something you guys can’t see. The difference between mechanical singulation and 3D unstructured bin picking, meaning kind of like what we have here just parts dumped in a bin no staging what so ever, is that bin picking maxes out on capability. You can handle multiple parts, you don’t need to worry about you presenting parts you know in a very specific way but mechanical singulation is better in all the other ways right. So you don’t need a lot of expertise like imagine designing an egg carton, apologies’ if anyone out there designs egg cartons I don’t know what goes into it but imagine designing egg cartons seems pretty straight forward, easy of use yea you put the eggs in them or your parts, initial investment probably a pretty cheap way to fixture your parts right. Bin picking is none of those its much more sophisticated, you need more expertise, you need to be savvy with not only robotics or, I say you, whoever is implementing the bin picking system to be savvy not only with robotics but also machine vision and also how to handle almost total unpredictability with your parts right. 

 

Why is bin picking so difficult?
There is much you can’t predict on a technical side. Imagine this is your part, this is what we have, with fixturing you put this in a fixture and you know exactly where that part is and all degrees are free right. If I just throw it in a bin I can’t predict any of those degrees of freedom so I need a camera to look at what’s going on there, do a bunch of math, and basically pick the best candidate. Now on top of just how your part is oriented it can be blocked by other parts, it can be entangled with other parts, you end effector can be such that you are going to collide with the bin and other parts um you can’t because you cant predict how your parts are going to be oriented you also cant really predict how your robots going to end up right so without a lot of testing you can’t say for sure that your robot isn’t going to reach so ridiculous singularity in trying to reach a part. So that kind of in a nutshell is what’s difficult about bin picking. I’m going to pause here real quickly just to take a question if anyone has a question.


Question and Answer

Q:  How many different parts can a system like this handle?
A: One of the really nice things about bin picking is that it can handle multiple parts. A fixture is designed for one part. With a bin picking camera if you tell it, as long as you have a 3D file of your part, the camera can handle more than one. Now what’s the upper limit? Honestly from the technical side it’s limited to the storage space on your hard drive where you are storing stuff. In a more practical sense it really depends on your tolerance sitting there programing. Because like, for example, this part on top of the cad model I need to teach the camera how I can grab the part right. So with this part I believe we have about eighteen or twenty different approaches set so each of these tabs you can it this tab can be approached four or five different ways the way we have it. This one a little less, this side can be grabbed so each part you want to teach or you want to bin pick requires kind of that level of attention. Pick up the more time you are willing to spend on a part the more successful your bin picking or the more robust I’ll say your bin picking appointments will be. To answer your question, the upper limit is really how much effort are you willing to spend to iron all this out. That’s actually one of the main benefits of bin picking so I appreciate that question.

 

Return on Investment
I have talked about how expensive bin picking is but how does our system with a cobot and a bin picking camera actually get you a return on that investment right. Well one of the major factors that any robotic deployment gains to mitigate is the labor cost right. People require money to work thankfully but the average burden rate so this is not, in your an operator position you have your hourly wage right which seems like a good way to sum up what that person costs but the fully burden rate to take into account training, onboarding, insurance, benefits, everything like that the average, a good number for the average manufacturing facility regardless of the skill level required for the position which is interesting is the fully burden rate which is between $22-25 an hour. Really, the true cost to have an operator for a position is in that ball park. And like I said many things contribute to this burden rate. Labor cost is the number one metric. It’s the number one thing that my customers tell me that they want to save on when we talk to them about robots. Now I have mentioned an hourly rate. Especially on the labor cost when you talk about multiple shifts, two, three shifts, your ROI becomes very rapid. So on the labor cost side, if you run multiple shifts definitely look at what your labor is costing you. Throughput again, especially with multiple shifts, robots scale really well. The benefits of a robot scale really well at your volumes increase. Your customer says hey next year I want to produce three times this amount, well the easiest solution is to throw three times as many people at it right, higher as many people as you can to make an increased capacity. But one robot, a robot doesn’t take lunch breaks, it doesn’t take any other kind of break, it doesn’t complain right, it never needs a day off, it doesn’t need food, um it doesn’t need performance reviews, oh well I guess you engineers do but it’s a different kind of review. What we’ve found is that even if you go in and evaluate an application and the operator is just flying through this process right and he’s saying well the robots cycle time is going to be a second or two slower than that so maybe the ROI’s not there. But when you consider that the robot is a consistency powerhouse over time often times even within a single shift the robot has outpaced a human operator in terms of cycles completed right. 

 

Ability to easily handle changeovers
The nice thing about bin pick specifically if you have a bunch of change overs say you need to fix your parts for the robot that eats into your throughput that eats in that contributes change over time. With a bin picking system when a bin is empty remove it put in a new bin and that is your change over right. Especially for high value parts the less time you can spend on non-value add touches to the part the higher your ROI is going to be. We’ll talk about high value parts in a few minutes. 

 

Flexibility
With a single bin picking system with a robot can handle can handle not only with multiple parts process as well. We’ll talk about multiple process later but mechanical fixturing and feeding I won’t say I it’s really right to say it cannot do this it’s much harder to do it with mechanical fixturing. Especially with something like bowl feeding, bowl feeding is great it’s great technology, its mature technology but when you talk about processing the tooling cost required to handle multiple SKUs with a single bowl for two or more SKUs it becomes unfeasible very quickly right. It gets much much more expensive. 2D vision, again, another great technology but it can depend on part symmetry quite a bit so with our part here this is a low voltage outlet box by the way you might have seen this if you work on your house with a 2D part or for 2D vision it might not be a great candidate because it has so many different way it can rest on a surface right. So 2D vision can depend a lot actually on what your part looks like. The flexibility of a process or of handling multiple processes specifically it lets you bring automation to where a single system might not be sufficient right. So maybe you have gone throughput on this process but every once in a while you run a different part through it. Well, if your tooling cost made that inaccessible for you maybe bin picking system might be a great candidate.  I said the thing about non-value add touch but every time you touch a part just to prepare it for a process it’s really hard to add value that way. So when you can take a single system and move it where its need this has been the big things about cobots in the last couple of years is that you can redeploy them right. Your ROI tends to scale rapidly. Change over, dictionary option I’ll actually save that for near the end if there is a question on it cause it’s kinda easy to nerd out on that a little bit but its basically our camera stores all the possible orientations of your cad file as a dictionary. When its presented as a new part orientation, it scrolls through that very quickly to find a match rather than calculating the part every time. This obviously minimizes change over time you need to well bin picking in general minimizes change over time. You don’t need to do parts specific fixturing anymore. And of course this is probably pretty obvious to you guys less change over time means more time in production, the less time your system is sitting idle the more time it is actually working for you. 
 

Reuse existing bins
Retooling, so now you may be able to use, you have parts in bins already - plastics people I am looking at you - you may be able to use the bins you already have right and especially for for smaller shops, smaller businesses that don’t have necessarily thousands and thousands of the same bin, replacing what bins you using its likely a lot cheaper than having a hundred different types of tooling custom engineered and fabricated. 
 

Redeployment
Removing the tooling requirement also removes the need to create a new process to present the parts to the robot. If your system, again redeploy ability, if the system is deployed to another process later those parts probably don’t need their own presentation method its just programing the robot and the camera at that point. So the tooling cost are significantly less at that point on a bin picking system. A major assumption of that is you don’t radically end of arm tools on the robot right but talking about part presentation there. And then retrofit the cobots are meant cobots in general are meant for mimicking human capably right so this system for using a cobot and a camera that kind of identifies parts using the same logic for humans by humans right. Using the same logic that people do when scoping parts out in a bin it’s a really great way to see what parts can be retrofitted instead of creating this whole new production line or a new process or having a new machine built you can just say okay there’s this person picking parts out of a bin all day, roll the system in right there just kind of bolt it on like I say in the slide right there. 

 

Safety considerations
If anyone is out there that I have come and spoken to you I probably have kind of tongue and cheek said if you put a chainsaw on the end of the robot it doesn’t matter what type of robot it is, its your application that’s hazardous not the fact that it’s a robot right. So with that disclaimer in a lot of applications you might not need any guarding at all and in other applications you might need an area scanner or a light curtain, tech or human presence. In a lot of other applications, you might need an estop outside of the robot work envelope right. But the collaborative robot the robot force and speed limited mitigates the need it solves a large part in your safety puzzle in deploying a robot. 

 

ROI Conclusion
The take away from this slide is that a retrofit almost by definition tends to cost less and it deploys much more quickly than a full machine build or a production line so a production line over all sorry. In value add, this is tangible if you are calculating ROI you probably have trouble putting this into a spreadsheet somewhere but people, again image picking parts out of a bin for eight hours that’s your job, that’s more or less unbearable and that’s kind of honestly a waste of brain power on a person so automating dull tasks like that. You can tune out and still do a task like that. Automating that task, unless you take a person and let them do something that adds more value to your organization. And again it’s hard to put that into a spreadsheet but from an intangible standpoint it the fact still remains the less people you need to do these repetitive, dull tasks, the more capacity you have for more complex task that people are actually good at. 


Question and Answer

Q: How would I interface a system like this to my machines? 
A: A very common answer to most questions in a presentation like this would to be it depends. And I’m sorry that’s exactly what I have to say in this case but it really depends on the interface of your particular machine, what kind of communication you’re machine protocols it can handle. On our system the component that would be reacting completely with you machine would be the robot and on that side what we got is a pretty good degree of flexibility. It can handle ethernet IP, TCP, mod boss, discrete digital I/O if that’s what you want to do. There are many options. 


Top Five Reasons to Choose Bin Picking

  1. When your part presentation can’t be changed up stream. I’ve had a lot of conversations where we say okay this is pretty straight forward but my parts are dumping out in a bin upstream and it’s an extruding machine for example and I can’t really slowly down the process for whatever reason I’m constrained upstream and I can’t change how the parts are presented without having an intermediate step without literally having someone spend time staging the parts right. Bin picking fits in really well there. 
  2. When flexibility is more important than cycle time. I haven’t talk about cycle time so let’s do that really quick. Cobots like I’ve said are meant for mimicking human capabilities. So there are certain situation where you have a cycle time of maybe a second or two, not really a great ball park for a cobot to be in. Every once in a while your stars might a line perfectly and you can get away with it. In most cases a good rule of thumb is maybe four or five seconds for a collaborative robot and actually with the amount of data and numbers the camera needs to crunch every time it scans the bin a couple of seconds is good there too. So, if you can afford to everyone to four seconds to figure out where the next pick is going to be that is great, you can actually do things. You can be clever about the programming it like when the robot picks a part and moves out of the camera field of view to do whatever the camera takes the part then and cuts your cycle time significantly to do that. Main take away from point number two is that if you need to be super consistent with you cycle time so you know I need my cycle time to be anywhere from three point five to three point seven seconds or even honestly six point five to six point seven seconds right. You’re not going to get that type of window of cycle times with your bin picking system just because your part density can change, part orientations can change, certain certain bins of even the same part could take a fraction of a second for the camera. You can certainly get within a window but cycle time is not the something that we chase with a bin picking system. It can be optimized but a certainty flexibility is the name of the game here. 
  3. High part volumes ties in part presentation up stream. The higher part volume I mean depending on your process the higher the part volume the less likely it is that you can afford the time to stage parts individually. So, at the very least vision becomes an attractive option whether its 2D or 3D with high part volumes. If your high part volumes are high enough the most cost effective is to just dump your parts into a bin to feed them into your next process. That’s where bin picking obviously really shines. 
  4. Multiple SKUs: Each SKU mechanically would need custom tooling. One bin picking station could handle several. As we’ve discussed earlier the upper limits is not really determined by you not technical limitation in a practical sense. 
  5. High value parts, we’ll talk about this in a little bit, if you just with automation in general if you have a ten dollar part you are producing ten of and you have a one dollar part that you are producing and you need ten of and you have the budget to automate one of those, which one is going to pay for itself quicker. It’s that ten-dollar part right because every part you complete it’s a bigger chuck of value that you move through width. So high value parts tend to pay for themselves very quickly or just in throughput and especially so with bin picking. 

Bin picking sweet spots
I’m going to go through a couple of process sweet spots and then industry sweet spots. Starting out with the processes is collaborative applications, like I have said cycle times of a cobot and the camera align pretty well already. Typically, in a lot especially in applications like a simple pick in place or machine tending where you are picking up parts not saw blades right but like picking up parts and feeding them to a process in a lot of cases no guarding is which makes it you know which not only makes the robot redeploy able but the bin picking system as well. Programming is a big one and its we get lost a lot of times and oh I don’t need to invest in a bunch of new safety like I would with another robot. Think about programming though right bin picking is optimizing it requires a lot of really hands on troubleshooting, debugging, looking at what’s going on. If you have to negotiate a cage right. So if you have an industrial robot we’ll say and you are trying to bin pick with it great you might have faster cycle times but anytime you want to see what’s going on through the programming process you’ve got to open the interlock right or some kind of guarding, robot powers down that’s a safe way of doing it. you’re there’s a lot of lag time in your programming that’s just not really with a collaborative robot because the robot can be running when you are right next to it and you can see exactly what the robots trying to do versus what you want it to do. So, programming is easier, and it just translates the system as a whole translates very well from how people bin pick. High volume with mixed products so the kind of the rough cut off I put here is three plus SKUs and that specifically is the point where bowl feeding ceases to become cost effective over bin picking right. Two SKUss, you could make that work when you get into three or more unique parts buying individual bowl feeders for each part, the number you have looks uglier and uglier right so high volume with fixed product your part presentation cost is going to be much much lower if you can bin pick all of that with the same system versus honestly even nested trays if you have that many products. High volume gives you faster ROI. That’s kind of coupled with throughput as well as labor costs. The robot will give you over time a much larger throughput just through sheer consistency. 

Repetitive motion injury risk
Repetitive high-volume tasks present a great risk of injury. How many of you guys type all day and now you know you have that pain along the back of your hand right. I see this with design engineers that do a lot of 3D modeling, they are clicking the mouse and moving it around all day. All the tendons through here hurt right. You’re not even picking up parts, you’re not even picking up you know a fraction of an ounce you’re just moving repetitively. High volume means you have all that motion the risk of repetitive motion injury is much higher for high volume processes. I think we’ve seen estimates of up to $80k per instance of lost time. Repetitive motion injuries so it doesn’t it’s not present everywhere but it is there and when it happens it can be a big savings to mitigate all that.
 

Routine tasks
People don’t want to do dull task and especially in the current job market if they see something better that’s you know even marginally better employers are having a lot of trouble with finding good operators that will do something that’s not the most pleasant thing to do right. 

 

Multiple bin recognition
Multiple bin recognition is something I haven’t talked about multiple parts, but the underlying assumption so far has been that I have one part here and then one part over here and then one part over here and then all the way on the other side of my plant I have a different part I can roll the system up. Kitting involves multiple parts in the same location right so one of the really cool things you can do with our solution is you can actually have different parts in different bins in the same field of view of the camera and the camera can take one snapshot and queue up multiple robot pics on multiple parts. Bin picking looks awesome really quickly in this scenario because again depending on the size of your kits, how many parts you have bowl feeding might not be an option but from a cycle time perspective not needing to do a separate scan of each different part you have saves a lot of cycle time there right. So, needing only one camera snapshot to capture all three bins recognize them and then tell the robot exactly what to do across all of them really makes an attractive sweet spot. If you have this feel free to give us a call, we’d love to talk to you and then high-value product ROI and break-even points scale very rapidly with high value like my one dollar versus ten-dollar part example. In almost all cases that I have looked at its been well worth the time to do bin picking or make that next level automation investment.

 

Consistency
Another kind of bonus here with automation is that when you automate something almost by definition you make the process more consistent right. That lets you iron out inconsistencies which typically means your quality goes up which typically means your waste goes down not automatically but the consistency lets you tweak your process to decrease the waste and for a high value product all his unnecessary waste can add up very quickly so if you can automate your process make it more consistent waste less it becomes an attractive option very quickly right. 

 

In what industries would bin picking be appropriate?

  1. Medical devices: Medical equipment is a sweet spot for several reasons they typically have high value product. In my experience it’s a lot of manual processes and a lot of times it's not that you know, something's being assembled. It's not that the assembly process itself is super complex. Historically for whatever reason there hasn't been a lot of Automation and all the parts are presented entrees or bins or or whatever. So each assembly process in an unstructured fashion. What are the bonuses for medical devices specifically? Is that there's this huge huge huge emphasis on serialization and logging everything about your process. So if something happens out the feel it can be traced back to the raw materials or something right like to which I'm sure you guys especially in a manual process writing down serial numbers back side or batch number Lot number. Whatever right automation can do the seamlessly for you and not only that it greatly reduces the chance of some kind of data logging error. 
  2. Plastic: Most plastic manufacturing environments are not operator right there at least two are either hot they run pretty hot.There's noise everywhere. If you're a if your job is to load a mold. For example, you've got the giant machine in and out from your face all day.This especially with the high turnover that almost all manufacturers are experiencing Plastics are having a really tough time because just it it's not pleasant for anybody and that's not so much a knock on the environment. On top of that you got most likely you got lots and lots of unstructured Parts all over the right because Heating and Cooling of your product is super tightly controlled if that machine stops for any reason everything comes up and you're looking at you know. Typically a day or more of downtime just to get that machine started back again. So you can stop at which means you probably don't have from extruder is as it comes out, right which means whatever size that's just cool places them on a conveyor the end of which we're out and the drop earlier, right? There's no real good way to predict exactly where that part or how that Parts in a y on a conveyor belt conveyor belt dumps the parts into event so over molding processes as well. Everyone that I've seen it involves an operator with you know, all the inserts are substrates in bins around the molding station and they just pick one place in the mold process happens Thompson together another been over mold opportunities right there. So Plastics is a sweet spot for many different reasons.
  3. Automotive is a sweet spot for Automation. In general Automotive knows the production process very well, as much as the operator. I said this a few times if your cheapest possible automation solution no matter what or seconds camera itself. Stuff everywhere in the processing time something like to division might actually be sufficient or you could do something clever with gravity feeding them that we Ransom perfectly but I'm saying this is our spot because if your parts are at least partially aligned, you may be able to do it more efficiently

Industries not suited for bin picking
Coils and springs: Again, bin picking is super easy for people right but it's hard for robots. If you had a bin of Springs in front of you that you could pick one out of the without another one. Even if you were limiting yourself to one hand like a robot,  the part repeatedly get those parts.


Question and Answer

Q: what parts can be picked and can the system be used to recognize clear plastic parts?
A:  There's a certain size limitation rights or camera tends to handle much smaller parts be able to handle in general. There's no hard-and-fast answer to what type of Parts you got sheet metal parts. We can handle it. Plastic parts, we can handle it. A lot of it depends on the end effector and how the system ties into your existing existing processor machine. Clear plastic parts can be problematic because they rely very heavily on how how well the light is reflected back. The best way to answer your question is the only way to really know is to send us your parts and we'll pass it on the camera. That's the case with honestly any Vision system. You should always make a vision test on the parts regardless of what the specs are. 


Interested in learning more? Take a look at robotic bin picking solutions offered by Cross, or contact a team member to discuss your application!

 

Contact us today!