Robots as Design Interfaces—Toward New Processes Beyond Mass Production

JOHANNES BRAUMANN: So good morning, everyone. Welcome to the session robots as design interfaces. We want to look a bit beyond using robots for mass fabrication, but towards mass customization. So my name is Johannes Braumann. I'm one of the co-founders of the Association for Robots in Architecture.

I am university professor at UFG Linz. And I'm also the main developer of KUKA prc, the software you're going to be seeing today. As a jet-lagged European, for me this is now the best time to hold a lecture. But I really appreciate you coming here quite early. So thank you very much. Ideally, I think we would do questions at the end.

However, if you have a really burning question, just raise your hand signifying, then we can do this also in between. OK. So I want to kind of showcase the developments that have been happening in the area of robotics within the creative industries in the past years. And then, also show you how these are now also transitioning from the creative industry to other industries that are also utilizing these methods.

I also want to point you towards the handouts that you can find in the download section. So the handout, on the one hand, contains a link to the software. So if you're using Autodesk Dynamo, you can actually immediately download the software and try it out for yourself. So we actually would really like you to try it out and to provide us with feedback.

We tested it on our PCs. But maybe we find out it only works on ours and none others. So this is a realistic risk. We're really looking forward to feedback. The PDF also contains step-by-step instructions just for really the essentials of getting started using this kind of visual programming with robots. So today we'll be having more of an overview.

But this should help you get started. We're also happy to provide some support. So all the information is actually then in the PDF. Finally my presentation-- we're not just going to show projects that we did ourselves, by Robots in Architecture or from our universities, but also from some of our partners that are using our software. So actually in the handout you have links to all our partners.

So if you believe that some project is interesting, if you want to have something-- robot 3D-printing cocktails, if you want to look into high-end timber fabrication, then, of course, you're very welcome to contact them. And we're always welcome to put you in touch. Because sometimes it's easier if you get the contact directly than just going to office@merktimber or whatever.

So please just get in touch if you could then-- any questions. So, yeah, but before we get ahead of ourselves. I think the main question is now the beginning what kind of robots we're actually talking about. Because robots can basically be anything. So just today in the morning the post-- there's some new videos from Boston Dynamics with a four-legged robot running around and acting more or less like a dog.

And, yeah, for us, we're really thinking about industrial robots or robotic arms. You probably know them from the automotive industry. And maybe you visited one of the automotive construction plants. And in the automotive industry, the reason why they really like to use robotic arms is that it's just much more affordable to take a universal machine like a robot, equip it with the tool you need for that particular operation.

Then you program it. And then the robot is going to perform that operation over its entire lifecycle, which usually is like, I don't know, five to seven years of 24/7 operation. This works really well for them. But the thing is really that this is mass fabrication. So the robots are programmed once with a comparably high effort and then they keep on running and running and running.

For our purposes, we're not so much looking towards mass fabrication. We're really looking to its mass customization. So maybe you've seen the Forge keynote yesterday where they also referred to this idea that you want to personalize, customize. You need to adapt products to certain conditions. This might be high-end architecture fabrication where you want to cover a free front surface with panels in a very efficient way.

This could be also providing customized product that fit the requirements of the person who ordered it. We got two robots quite many years ago. We, in that case, is my colleague Sigrid Brell-Cokcan, who is now professor at RWTH Aachen in Germany. And in about 2007, I would say, Sigrid got the funding to buy a CNC machine. So this was a research project with the geometry department.

And the idea was to look into mighty axis fabrication for architecture. But they had a budget of around 100,000 euros. And there was really barely enough to get the cheapest five-axis milling machine at that point in time. I think it has changed a little bit. But at that time, this was just the price level that we were at. And that machine wasn't really trust inspiring. Even during the demonstration it broke down.

So this was, of course, a bad start. We figured we need something reliable, but at the same time, we want to have something innovative. So just getting a three-axis milling machine and saving 50,000 euros wasn't really an option. So at that time, KUKA approached us and told us that they have a product that would allow us to take g-code meant for five-axis machines and then use it on an industrial robot. That was called KUKA CAMRob. Now it's called KUKA CNC.

And that was actually first a reason to say, well, we're going to take a risk. We're not getting approved CNC machine. We want to look at robotics. Of course, it also helped that, with the idea of research of course, we wanted to go somewhere where there isn't that much existing projects going on here. So we got the robot. We bought the software. So all of that somehow fit into the 100,000 euros.

And that is the kind of workflow that was provided to us by KUKA and the CAM company. So basically, if we wanted to mill with the robot, we had to first make a geometric design in a software like Inventor, Rhino or maybe now Fusion. But that didn't exist then. Then you would have to export it in a generic format, like IGES or STEP or something like also DXF if you really want to.

And then you put it into your CAM software, into computer aided manufacturing. So this would be, for example, today PowerMill, if you want to have an Autodesk product for it. From there on, we again had to export our five-axis g-code to the robot simulation environment that would show us how the robot would move in order to fabricate it.

And only then we would be able to send the file to the robot. So that's a workflow that actually works pretty well for industry if you want to do individual prototypes. So we work with partners who do large-scale stone sculptures. And this is a workflow that's perfectly fine. However, for us, all of that was quite complicated.

Because when we were looking into the milling software and then we see that there would be tool reachability problems or some problem with the geometry, that it didn't input correctly. We had to go back to the geometric design. Or even worse, if we ever realized in the robot simulation that the robot cannot reach that point, we might have to go several steps back in order to be able to solve that issue.

So that's not so much of a problem if you're very experienced CAM or robot operator. So at some point, you get a feeling how all of that works. And you can go through that workflow quite well. But for us, the problem was we were looking at education and we were looking at people who didn't have any previous CAM experience.

So actually, in our teaching, we spent the entire semester teaching our students software that they most likely would never ever use again. And the robot was really like the part at the very end of it. So we didn't spend the semester working with robot. We actually worked 3/4 or even more of our time only on software. And then at the very end, we had a few days of fabrication.

And that was not really what we were looking for. And at the same time, at this stage there were also developments that were looking more into programmed geometry. So not so much just kind of traditional way of drawing something for fabrication or 3D scanning something that you fabricate, but that you actually start looking into fabrication and design as kind of one process.

So that was one of our first-- some experiments that we presented at the university. So you see that basically, in this case, we don't only design geometry. We actually design tool paths. So whenever the tool path changes, the geometry changes along with it. So basically, it has got both of these aspects in the same kind of project.

But in order to fabricate it, we really had to kind of dumb down the parametric geometry that we have. So we had to, again, save it as an IGES or whatever, bring it to the milling software where we actually had all the information already that we needed for fabrication. And so this was for us kind of a problem of course. And the other thing is, it's not very difficult to really generate code for the robot.

So that your just write your machine code for the robot, this is something that you can achieve quite quickly. But the tricky aspect is really how to get the simulation right. If you don't get the simulation right-- that's a project where we had eight liters of glue for a student project. And they programmed all the position in the bucket.

They didn't simulate how the robot would then from the last position go back to its home position. So, yeah, that happened. Anyway, let's not look at it again. The other way, of course, to approach this is that we just stay, well, we want to script that, separate scripting language. We could do this in Python or whatever. So we script our constraints. Then, we put it into the simulator. Then, we debug it. We change it again.

But this was also for us not really where we wanted to go. So for us, we wanted to have something like on the right side. That we have one unified environment that contains the geometry but also contains the fabrication. With the idea being that whenever you do something to the design you're going to see how it affects the simulation and how it affects the robot, basically, in real time. So just keep a few years ahead, this is how that is looking now in Autodesk Revit and Autodesk Dynamo.

So this is just a very simplistic sample. So you draw a curve. Then, you basically divided into different segments. And then, only at the very end, you plug in the components for the robot. And you can immediately see how the robot is going to react when you fabricate it. So here you already see that there's a problem. So the robot has certain limits for its access. It's going to collide with each other. So you can't do that.

But now you can just change the curve and the fabrication updates along with it in real time. So the idea really being that rather than having to know all of this by experience or by a very lengthy trial and error, you have this immediate feedback. So you need to think of it a little bit like digital photography. You take a photo. You see what the design is. You can react on it.

So you can work with this in a more intuitive but less systematic way, which is good. But of course, if you have this knowledge, you can also very quickly implement that. And the nice part about the visual programming, of course, is that we don't need to expose people to all the logic that's happening in there. So basically, we have a component that has a certain set of inputs and a certain set of outputs.

And you don't necessarily have to know what's happening inside. For you it's just important to understand if we put that in, what do we expect to get out on the other side. So now to go back quite a few years again. We started to do workshops. Because we figured this would be a method and an approach that would be relevant to many people. And we're quite astonished at the very beginning.

So this was our very first workshop actually. We got many people from industry also universities who were also looking into these kind of problems and were looking to solve them. But we also figured out, at the same time, that if we're just two people doing workshops driving around, this is not really sufficient to have an impact on the creative industry or even at industry.

So for us, it was the impact to say we need a platform. We need to have a network of users of these robotic arms, of industrial robots of these processes so that we can exchange knowledge and also distribute it much more efficiently than just the two of us. So in 2011, we founded the Association for Robot Architecture-- at that time as a spin-off from TU Vienna-- on the one hand as a network of robot users but on the other hand also as a research institution.

So that has grown quite a lot since then. So this is a map of the members. So we now have about 100 university members worldwide. So you see kind of the hotspot with a current list of most development and also 50 commercial members, people who use our software for commercial purposes. And this really also spans a huge range of applications.

So this is not just in milling or not just in 3D printing but also on all different kind of levels. So on the one hand we have very large companies who have machines that can process 50 by four meter elements of wood, like in this example. MERK Timber will use that kind of approach. Anyway, so this is how they, for example, work.

So this is really high end fabrication for architectural purposes. They did projects like the Elephant Hall in Zurich, if you know that, and also some very interesting projects in Barcelona and now also even more in Switzerland. And on the other end of the spectrum, you don't have this kind of mighty 100,000 euro installation.

But you really have people who go on eBay, search for KUKA robot, and basically buy the cheapest one. And then a very happy when they finally get to robot to do what it's supposed to be doing. So it is a group that calls themselves Silly Sods in the Mud. And they build boats in Great Britain. Otherwise, the range of partners really spans many industries.

So we have companies like Boeing and Adidas. We have companies like MERK. We have quite a few universities, as I mentioned. And, I think, are really key for this kind of efficient exchange and advancement innovation is really that we have so many different areas where people are working in and this kind of cross disciplinarity is really a turnkey for successful development.

As a network, of course, it's also important that we promote this via the internet. And so these are just some of the facts. So depending on which industry you in, this might be larger. This mighty very small numbers. But if you see this in context, if you think about how many people are using robots in this kind of purposes, where on the other hand, have 15,000 downloads each year.

And the Springer Link-- this is like a scientific publication house and having 140,000 times people access these kinds of research papers shows that there's definitely an impact. But of course, having an online is just one part of that. A big part is always about meeting offline, basically, in the real world discussing the kind of problems and having an outlook.

So every two years, we organize conferences on robotic fabrication in architecture, art, and design. We started out in Vienna 2012. Then we were at the University of Michigan in 2014, at the University of Sydney in 2016. And in 2018, we're going to be at ETH Zurich in Switzerland. And one part of that is conference panels. So basically what we are having now, lectures.

But another big part of the robot conference is that we have three days of very intensive workshops. So people not only talk about these developments, but can really experience it hands on. Because as you definitely know, if you make a video, it's very easy to cheat. But if you do this hands-on, you really see the issues and you can learn much more from it.

So this was a project from the University of Michigan at the conference 2014. This was from Bot and Dolly. So maybe you know them. They have been bought by Google. So they experiment with this kind of skill transfer from motion capturing to the robot. Or this was ICD Stuttgart, who have developed this really impressive carbon fiber structures and also showed on-site how this actually worked with two robots cooperating in a smaller scale.

So if this is something that's interesting to you, you still have quite some time. So this is going to be in mid-September 2018 in Zurich and Switzerland. And just to kind of explain how we're internally made up. So initially, we had the Association for Robots in Architecture. We also have our Robots in Architecture Research LLC for commercial projects.

And then we have a two university chairs, the University for Arts and Design in Linz, where I've got my professorship. And my colleague Sigrid is Professor at RWH-Aachen, where he's got a chair for individualized production. And we're then also very much involved in the area of construction robotics. So in Linz, a very big part of our work is done with ARS Electronica Center, which is one of the largest digital media centers in the world.

So we work also on projects that don't really involve physical fabrication, but can be more installation. But at the same time, we work with craftsmen and SMEs to really make robotic processes available for them. At Sigrid's chair in Aachen, the idea is to look into large-scale architectural production, the digitization of the construction site, and really bringing robots on site.

So we closely work together. But we try to kind of arrange projects so that we can best support them. So you're already seeing quite a few projects that we have been done with robots. So maybe the question is, what can we actually do with robots? And the answer is a bit flat. But basically, you can do nearly everything.

If you feel like you want to feed your baby, which I don't recommend, but if you want to do that, it's definitely possible. This was for digital media project with a German broadcasting company called Homo Digitalis. So we took very good care of the baby and the robot. So everyone left happily. But basically, whatever you do, it's not enough that you can just program it. You really need to know about the material processes as well.

So if you want to use a robot to structure stone like this, then this is not just programming. You really need to know how to do this by hand. You need to understand how all of that is working. There is, of course, some software that supports you. We already saw the milling software. And for example, here we have Autodesk Fusion.

And we can also deal with this kind of information in our software environment, turn this into robot code simulated. But sometimes you have special purposes and special requirements, that you really need to do something that is beyond the scope of industrial software that, of course, is made to cover the most common requirements. So this can be, basically, a big range of things.

Here we had a project that was done by two artists, Neugeauer and Kolldorfer for Red Bull. So this was a monumental 17 by 23-meter aluminum sculpture for the Formula One racing track. And even though you might not really believe it with Red Bull, the main constraint that they had here was actually the price. So they drew all the forms in 3D. Then, they send it off to manufacturers. And then, they got quotes back that were extremely high.

Probably you know that if the manufacturers are not 100% confident in your capabilities, they would just add 50% on top of that. So this probably also happened there. So for them the idea was, well, let's just do it ourselves. They knew that we were working with robots at TU Vienna. So they got their own robotic arm, bought it used in Germany for 20,000 euros, put it on a truck, set it up in the lab.

What they then actually realized is that the robot itself is of course not enough to fabricate. The idea is you take a 3D-model. You put it onto the robot. You click play and then it would build the piece. If you worked with CAM software, you know that this is not the case. And especially the CAM software used to drive the robot would cost more than the robot itself.

So basically, we worked with them to create this kind of customized interface that would allow them to perform their operations that are needed for them. So not like a universal solution that you get out of CAM software, but something that is very specific for their purposes and allowed them to do that. And the special part about this is that we, of course, didn't have an interest to spend several months away from our own work, away from whatever to fabricate a huge monumental bull for Red Bull.

So the idea was really to have this kind of knowledge transfer. So for just four days we worked with the guy in yellow, who was the son of the artist, who was basically Shanghai'd to work on his father's project. And over four days we taught him how to use the robot, how to use the software. And then they were really able to fabricate it.

And what's special about that is that the guy didn't have any previous knowledge about CAD or CAM. So he hadn't ever used any kind of CAD software. Before he had studied tourism. But still the kind of four days were enough for him to kind of understand the essential parts that would allow them to really fabricate this by themselves, then cast this kind of positive foam parts into aluminum. And then set them up on site.

Here with Zublin-- that's of course, the other side of the spectrum. So this is very high-end architectural fabrication. This is for a train station in Germany. So these kind of free form columns. So you see there are several of them. They all look differently. And each of form work for these kind of structures up here, the columns, again, consists of these kind of individual elements.

So if you do this traditionally, you end up with having-- I didn't count them-- 12 or whatever elements per column. Then you have so many columns. And you need to process each of them manually. But of course the thing is all of these elements are individually. So they have a different angle here. They have a different curvature. But the general topology is the same.

So in that case, the approach of Zublin was to say, well, if they use Grasshopper, they don't have to process everything individually. But they can extract that information from the parametric model, do the fabrication, and then send this to the robot much more automatically than before. Because as I said, they all basically work the same. You only need to be able to adjust the angles.

At the same time, the motivation for them was also that they didn't want to use kind of standard CAM software. So the idea was that if they use CAM software that all the competitors are using, they don't have a competitive edge. So instead, what they wanted to do is that they wanted to be able to take their own process and material knowledge, put it into the software, and then benefit from it.

And they're doing this through visual programming so that they can very easily adjust and modify that. So this is it just a very simplified example how that could work. So you have your free form out of Revit. It's panelized in into different panels. And then you can just automatically generate your files for each and every part. So you see they all look similar. In this case, they all have four corners.

But at the same time, everything is individual. Of course this is kind of slow scale. You could just do this automatically and get a bunch of-- hundreds of files. But this way, you can actually see what is going on. And you especially need this if you want to adjust to more individual parts. So this is a project by the Architectural Association at Hooke Park in London.

It's a research pavilion that they developed with actually found wood. So they need processes that are able to very quickly adapt to the local conditions, which in that case is, for example, the size and diameter of the piece of wood. And then of course, do this very efficiently for many pieces. So that they were then able to assemble this to one continuous structure made out of individual found wood elements.

And then sometimes you have projects that you do by hand so, for example, hotwire cutting. So many of you have probably done this before. So that's a company in Germany, Artis Engineering. They work in prototyping. And they need to fabricate things very efficiently. Because they don't do this in high numbers. They do this for a customer, maybe there's only one piece.

And if you do this by milling, you would have to cut away everything that's basically here. So this takes very long and produces waste and all other kind of problems. If you do this by hotwire cutting, you see in five minutes you cut through. And with this geometry, you can very quickly and efficiently fabricate it. Now the thing is there is no really a software for hotwire cutting.

So again, for them the idea was-- they actually built their own hotwire cutter, just out of aluminum profiles and a wire. And then created their own software tools. And then bringing that together, they had a completely new solution for that. And of course, then in visual programming, again, just a very simplified example. If you can also look at all kind of elements so you can check-- does the reachability work? Does it not work? Does it cause problems?

And some kind of applications you can, of course, solve by brute force. So if you want to bend metal with a robot-- In this case, to be entirely honest, it was just enough to have enough payload for the robot and bend it with as much as force as it had. So this was for a company called Tylko. They do mass customized furniture systems. So they have their own iPad app where you can adjust your furniture to your living room.

And they wanted to look into how they could work with metal. But then, for example, if you want to do knitting with robot-- so this is a student project from Poland. Then of course, it's not enough that you just know how to program it, but you actually need to understand the process. So probably you need to do this a few times by hand in order to then really get the result that you require.

And I mean, again, this is something that, just like hotwire cutting, we've been doing the hand. But of course, we can also do it on new things. So maybe we don't knit with some kind of wire, but maybe we use wire and resin to create structural panels. So this was by the IAAC in Barcelona at the visiting school in Moscow. And if you work with new materials with new processes, it's not enough just to develop the hardware.

As you see, you also need to think about how do you develop the processes. So for example, this is a new 3D printing process developed by an Italian research lab. And they had the idea of 3D printing like a silkworm. So he didn't want to melt plastic, but they actually extrude fiber that has been put through resin and then the blue light that you're seeing is then UV light that's used to harden it.

And if you want to do something like that, it's not enough just to take your standard 3D-printing software and use the g-code from that one. But you see that it actually has to tilt the tool. So you need to define a strategy how you can extrude that in efficient matter to do this with a new process. And of course, you also get new challenges if you go up in scale. So it's one thing to print in small scale. But printing in large scale has its own challenges.

So that's one of our partners, Branch Technology. They're based in Chattanooga in Tennessee. And they developed a large-scale 3D printing process. But this is not university research as you've seen before. But this is then really industrial research. So their company there, as far as I know, venture capital funded. And are really now bringing this out in large scale.

And having this kind of custom interface allowed them to create their own logic that their competitors don't have and then to build up on that, to improve on it. And now they're already working on very large scale-- like these kind of structures. And also, for example, involved in research. So this is done with Foster and Partners, a big British architectural office where they won the NASA Mars Habitat prize.

So they develop these kind of strategies-- how we could 3D print in the future to create structures on Mars. And sometimes robotics is not just about fabrication. It's also a tool for, I would say, just entertainment. So for example, this is a project that is a spin-off out of my lab. It's called print a drink. And it was developed by Benjamin Greimel with the idea of printing liquid within liquid. So this is not some kind of structural element. This is not 3D printing for construction.

This is really for marketing. So he offers this to companies. He puts at events his own logo into the cocktails. And then people can drink it. And it's actually resided out of a university course. So we had it with the concept of thinking up machine and food design. Do something involving food. But don't duplicate what did you by hand, think a it further what you can do.

And he developed the liquid in liquid printing. And then really was able to push that into a startup. So it's now very successful, has its own robots and offering this to companies both in the states but also, of course, very much in Europe. And here the nice thing really was that the robot was something that came at the very end.

So the semester was really spent on material research. So this is the kind of proprietary knowledge that you have in here. The robot at the end was a tool that was very easily added. And he developed his own deposition tool, which is basically it has some medical equipment put it into a 3D-printed environment. So basically the robot allowed him to have this kind of proof of concept that it works to show people, show investors, that this is something that's valuable.

And now he's actually building his own machine to do that but, of course, like this because in a bar, you probably don't want to have a robot standing somewhere. I mean actually yesterday in Las Vegas we were at the Tipsy Robot Bar, where they have a robot bar keepers. So this doesn't apply to Las Vegas. But usually you might not want to have a robot in the bars. And now he's building his own machines.

And yeah, sometimes it's not just about entertainment or fabrication, but it's really just but entertainment. So for example, this is a company in Australia called SEEit. And they actually use our software to create movement for a Robocoaster. So that's a KUKA robot that's specially certified that you are legally allowed to mount people in front of it. And they created their own virtual reality experience and then used our software to create robot movement that then works parallel along with it. And, yeah, it seems fun.

So this was kind of a wild ride through many different projects. So in the last five minutes, I want to give you an overview, an outlook on which direction the research is going, and also the challenges that we are facing. So we want to look at real-time interfaces, man-machine collaboration, and then also cloud robotics.

So real-time interfaces is something that might seem very primitive to you, because you're used to taking a printer, plugging it into your laptop with your USB cable, and then you can immediately stream commands, or print jobs to it, and it just works. With robot, we have the challenge that they're not really made for that.

So we can either control them in hard real time-- which is an issue to do, for example, out of Windows, so it's a bit technical topic, so we'll skip that a little bit-- but the other way is, of course, that we need other ways how to do that in a safe and reliable way to get commands from our very flexible visual programming system to the robot.

So there's only been some developments recently, and KUKA developed a system called mxAutomation. And mxAutomation is actually made not for visual programming, or not for this kind of normal programming at all, they actually developed it to have CNC machines, CNC controllers-- so-called PLCs-- be able to control a robot. So the idea being that an operator who knows how to program this kind of PLC, this kind of industrial computer, then doesn't have to learn a new programming language, but they can immediately program the robot along with it.

And we kind of adjusted the system so that we can not only do this out of these very specific PLCs, but that we can integrate this into our Windows-based, Dot Net-based robot programming library. So in that case, we only have this kind of code on our side dealing with the geometry, dealing with the tool path.

And then actually the robot side, all of that part is provided by KUKA. Which is for us very good, because then these are technologies that are very easy to hand off to partners. Whenever you do something that's very specific, that it's very specific hardware, it makes it hard to make it available to partners around the world. And you have to know that we are working with universities who, of course, sometimes might be a bit reckless. So they see some code on the internet, and they just start it. So that's a bit of a challenge for us.

So this is for us a safe way that we can communicate with the robot.

But of course, you need to think about how you can do this kind of communication. And in that case, we always have a command buffer. So this is basically some storage at the end that saves the commands that are sent to the robot, and then executes it. So we don't have to always keep on sending commands, as long as the buffer is full. But then we can create these kind of processes that then always change between evaluating, and then doing something.

So the robot with 3D scan our visual programming environment would generate the tool paths that it would do that, that it would go back again. Or it could automatically process many, many pieces based on real-time data.

So we first presented this at, actually, Autodesk University, but in Germany. Just some very simple examples here. And then again, we did a hands-on demo at the AEC Hackathon, also hosted by Autodesk in Munich. The kind of present purpose, utilizations, how you can use then real-time control of robots in their own design systems.

Maybe that was a bit technical, so let's go to a bit more fun process, man-machine collaboration.

I'm not really thinking about, man-machine collaboration in this very idea that you always have, robot and machine working next to each other. In that case, I'm kind of more thinking about what can the robot do, and what can you do by hand? So this is actually output of a project we did with the university for applied arts in Vienna. It was called robotic woodcraft.

We're looking into more arts-based application of robots. And this was one of the processes that was developed by one of our research assistants. And he had the idea that, he can very quickly manufacture this kind of ruled surfaces with a robot, because for a robot, doing this is very easy. You just make these kind of holes, it's done in five minutes. If you do this by hand, it always takes ages.

But something that you cannot really do by hand is this-- what you cannot do with the robot properly is the assembly. Because the problem is that you always basically program the robot at zero tolerance. So the robot is very accurate, the robot is that precise, you tell the robot to go somewhere, and it's going to go somewhere.

With wood we have the problem-- and with actually most materials-- that there are always some tolerances. So this kind of wooden roadster always bent in some degree. If the humidity changes in the air, also our wooden rods change. So that's a problem for us. And we always, again, had to do this by hand.

So for our research, together with KUKA, we wanted to look into ways, how we can get a robot to deal with tolerances. This also looks at the future idea of having robots on construction sites, where you also have this kind of problem that you might have a very accurate 3D model. But it doesn't really mean that the 3D model coincides with the actual physical environment.

So this was something we did for the Hanover fair in Germany, which is one of the largest industrial fairs in the world for KUKA, and we used a very special robot for that. So you see that robot up here looks a bit different. It's a KUKA LBR IIWA-- with IIWA standing for intelligent industrial work assistant.

Basically, they built the robot to be safe to use next to people. If you have a normal robot, and you tell it to go from here to here, and you're standing in between, you will probably get hurt if you're not quick enough. So that robot doesn't have any senses-- it's just going to go from here to here, and everything else, you need to take care of yourself.

That robot has got four sensors in every motor. So if it feels resistance, depending on how you program it, it's going to stop. But we wanted to use this kind of sensitivity a bit different-- so not only for safety, but also for assembly. So for this kind trade fair installation, we basically first captured some initial data, because we didn't want to have everything the same every time, because then it would be kind of faking it. So in this case, actually people took the robot in front, and then they captured this kind of curve.

And that curve was then the information that we used to inform the fabrication process. The next step was then for the robot to take the rods out of a magazine, so they are not placed very accurately, they are somehow thrown in. So this is not super accurate placement.

And then the first thing it has to do, it has to figure out how long it is. Because they're not very accurate, as I talked about. So actually-- that was cut away here-- actually what it did to measure that wasn't using some kind of optical sensors, but it just took the rod, went down until it collided, then it knew how long it is on that side. Turned around, went down again.

Then it knew how long it is on the other side. Then it had the total length. So then it could cut away the part that wasn't needed. And then it would assemble it on site.

And the assembly, again, works a bit similar, like we do it by hand. Because we looked into how you assemble by hand, and actually, just by looking at it, isn't really enough information to put something like that into a hole. So we usually get the rough information by looking at it, and then we assemble by feeling.

And we did the same here, but we didn't have a camera system. So instead we had our ideal 3D model that told us, this is where it's supposed to be. And then the robot, with the sensitivity, adapted to it in real-time. So it just meant this [INAUDIBLE] movement, and once it slipped in, it knew now it is at the right position and it can release.

So this is how all of that, then, looks in one entire process, so we could teach it the curve. Then it measures in the back how long it is, and then it assembles it.

So this is now not some kind of architectural structure or pavilion, or whatever, this is really a fabrication prototype showing how these kind of processes can be used. For example, in environments where you don't have optical sensors. So 3D scanners are great, but once it gets dusty, once there is big humidity, you run into issues, and then having these kind of processes is valuable.

So you can program that robot out of the visual programming environment. You can also-- actually, also have that robot on a mobile platform that's available in [INAUDIBLE], so you can have this very small robot cover very large environment. So this is going to be a very big part of research in the future.

And to close off, I want to look a little bit into the cloud. So we have now seen where processes where you use with your programming, where we use, for example, Autodesk Dynamo, to get processes from our CAD to the robot. And this really lowers the kind of entry level considerably. So you didn't have to be a robot expert to use it, you could very intuitively work with it.

But again, this is assuming that you know about CAD, that you are somehow feel comfortable looking into visual programming, even though it's only visual programming. But of course, this is not something how you could reach the end customer. So this was for us the motivation to actually bring our entire library into the cloud, and then work with the Autodesk Forge environment to visualize it in 3D in the browser, so we can actually make processes available to the end user.