Andreas Bihlmaier talks about modular robotics and starting a robotics company. Andreas shares how he started out in computers and later felt that robotics, through its combination of software and hardware that interacts with the world, was what he found most interesting. Andreas is one of the founders of RoboDev, a company that aims to make automation more available using modular robotics. He explains how modular systems are especially well suited for automating low volume series and how they work with customers to simplify automation. He also discusses how a system that can easily be assembled into many different robots creates an advantage both in education and in industrial automation, by providing efficiency, flexibility and speed. Reducing the threshold to automation will allow a new type of exploration of robotics and automation in STEM/STEAM education and in prototyping for industrial robotics. We get a personal, behind the scenes account of how the company has evolved as well as insights into the reasoning behind strategic choices made in product development.
More about Robodev.
This podcast is part of the Wevolver network. Wevolver is a platform & community providing engineers informative content to help them innovate.
Learn more at Wevolver.com
Per: Welcome to the podcast episode of Robots in Depth and this launch episode with Adreas Bihlmaier in cooperation with Wevolver. Robots in Depth is supported by Aptomica. Today, I’m super excited to have Adreas Bihlmaier here from Robodev and we’re going to talk modular robotics, my absolute favorite topic. I love all the topics in the show but I have this special passion for modular robotics. We are going to start where we always start. How did you get into the field? What was the first step of the journey?
Andreas: I guess as many later engineers I was always tinkering around as a kid with different kinds of hardware. Then I got into the whole computers, mostly gaming in the beginning but then the more and more I was sitting in front of the computer I realize you can do much more interesting things than just shooting down enemies in virtual worlds. I got into programming. I was developing a lot in terms of operating systems. I was working with the OpenBSD system. I contributed there and pretty much I was mostly interested in computer security and operating systems so not at all related to robotics. Somewhere at the beginning of my undergraduate studies I walked into two or three robotics labs which is hard not to do when you study in Karlsruher are there are several robotic institutes. I thought that is not really interesting. It’s still mostly software development because all belong to the faculty of computer science but it interacts with the real world and this aspect of the software interacting with the real world that really got me into it. I started playing with simple systems at home like Arduinos, ATmega controllers, stepper motors, really simple, also low cost, as a student low cost robotics. Then what really got me into robotics was that I started with a colleague of mine Swarm Robotics Project for high school education. We try to build a robot system where each robot was less than 50 euros. At that time I wasn’t too deep into entrepreneurship. One cost was 50 euros, the many hours we put into it and sampling it of course we didn’t count either but it was a volunteer project. We sew it together by hand, 15 robots that had a mouse sensor to drive on a two dimensional surface that could communicate with each other using infrared signals and had a bunch of other nice features. We had this kit of 15 robots we went around Karlsruher where I was born.
The children were really excited because they had simple tasks we gave them like let the robot drive around in rectangular meaning drive forward a few centimeters, turn 90 degrees, more forward again. Not really advanced programming but we made in one classic comparison between doing something like that with just interaction on the screen and having the real robot on the table. Having this physical interaction really got all the children excited and I guess I also stayed a child in that regard. I think it’s just fascinating. You have code that influences the physical world which also means that you have a back and it crashes you cannot just reset the process but something really crashed. That also kind of increases your game. You really have to pay much more attention to what you’re doing and the real world poses many more diverse challenges than if you are just in the virtual world where everything is probably defined.
That was kind of how I got interested in robotics. The second half of my undergraduate studies I did a lot of robotics topics and then also took my diploma thesis in self-reconfigurable modular robots. That was really an awesome project. It was simply a replicator project and I was closely involved in the last step so I wanted to do a PhD modular robotics. Unfortunately there was no further funding for projects in that area so I would have to leave Germany which was a plan but then they came a collaborative research center up between Karlsruher and Medical University of Heidelberg and the inter-disciplinary part of that really got me into it. Working together with surgeons on a daily basis but being a roboticist trying to give them system of which they can heal patients. That is quite a motivation. I did my PhD for three and a half years and during that time then Robodev came up but perhaps I’ll elaborate on that later on.
Per: Bringing robot to schools and having this physical thing move in your program. I’ve seen that this is so strong for the students. They really love the fact that they can program on the screen or they can program even on the robot. We’ve done this with the students in the age from four or five all the way up to university professors. Everybody loves the physical robot.
Andreas: It’s just different. We are humans. We are spatial entities ourselves. We are embodied as the AI or the cognition guys like to call it. It makes a difference whether it’s 2D graphics on the screen or in your own real world environment. We just react differently to that and why not use that to motivate children to go not just into robotics but general for technical topics for computer science, for mechanical engineering. That is also really an advantage of robotics in terms of education. You can do software. You can also do mechanical engineering, electrical engineering. You can even go down to physics, talk about battery circuits and so on. You can bring in a lot of different aspects with working with robots. That is also why I went to one of the Education in Robotics Conference was part of the conference in Italy in 2016. I think there should be even more drive behind this, even more political will especially because in the end somebody has to pay for it. Perhaps new models come up to make it cheaper for schools but I think it’s a great device having robots especially having cooperating robots. It doesn’t have to be swarm but the children do not just sit there on their own or in groups of two but over the course of the lecture they actually join in and build a team and that is even more motivation from my experience at least.
Per: I’m absolutely convinced that that is true. In Sweden we have added this to the standard curriculum, everybody should know a bit of programming and then everybody is looking for ways to do this best. I think that as we both agree there is a consensus that at least for younger age groups if you have this physical robot that does something it’s easier than to teach them. That is why I also love modular robotics is because this takes this a step further. You can’t only program the robot but you can build any robot that you can imagine. They can add to their learning as they grow.
Andreas: It gives even more possibilities what you can do with your one system. It’s still one system but you can use it in many different manners. You can put it together differently. There is definitely an advantage in modularity. You could guess that I’m a fan of modular systems.
Per: Everybody knows I am. Then you started to look at maybe starting a company. Can you talk to us about the time before and how you ran up to starting the company?
Andreas: Myself and two colleagues which I’ve founded Robodev, we realized that if you really want to bring this aspect of easy to use modular system to move this further along not even just to bring it to the market just to continue working on this subject we need to find proper funding for it and we realize pretty soon it won’t be academic funding. We have to go to the private market which means we have to go towards a product rather than just a research project that can then perhaps be commercialized later. Basically how can we reduce the effort it took us in our research for various projects. It was not only matic robotic but also service robotic projects that a colleague was working on. How can we simply reduce the amount of effort to integrate actuators with sensors with the whole logic control? Obviously I used ROS a lot during my PhD. That is great for the software aspect. It brings you a lot of drivers. You can control robot arms. You can read out sensor data and everything both sides kind of the actuation and the sensors are connected in the middle by a personal computer or a whole network of those. That was really nice to work with such a system but still it was a lot of effort to put together the mechanic or the electromechanical basis so you had your drive and you need to wire it up to your electrical cabinet and you’ve got your motor driver that needs to be connected to a PLC. Then you need to write some bridge to connect it to ROS or a personal computer.
That took us so many days and weeks and we only want to answer specific research question and obviously write a paper about it. We spent so much time with basic integration and this aspect how to package basic integration into usable modules. That was kind of what drove us towards what became Robodev later on. Just in parallel to our research projects usually done in the evening hours we were looking how can we build combination of software and hardware that gets us to running systems faster. Initially that was very much focused on electronics like naked PCBs to connect the motor to and software but the whole mechanical engineering part was lacking. Then you could connect the motor but your whole mechanic system to get the robot was still hard to come to. Then we fortunately could convince (0:11:01.3) then later on founder of Robodev was a mechanical engineering by study to also join us. Then he had a really good idea let’s use aluminium profiles as our basic structure and integrate everything into them so that you can use standard connectors. You have such a big catalog of off the shelf components in terms of structures.
Per: Connecting pieces, everything.
Andreas: Basically everything. You do not to go to your CAD software design something and flange for connecting motor to axis and have that milled and produce. You just go and take off the shelf components and basically screw them together. Also take them apart without destroying them which is an important aspect as well. That is what is kind of the beginning. We were just working on it as a side project. We intended it for research and perhaps also a hobbyist maker purposes. There is a wall coming up and this wall is ending our PhDs and having to find something that pays our bills. We look into are there other people that perhaps require such modular systems not only in research because that is a very special market if you only cater to researchers as customer. Is there industrial production of a similar demand for systems that are much faster to put together and put into production? We talked to a lot of people within manufacturing companies and everybody said yes, if you can give me something which I can automate tasks that are below the automation threshold, the volume is too small, the variety is too high, the product life cycle is too short that is definitely a business case and from thereon we looked at the market analysis for going with our system into the automation market what could be potential customers and so on. Then it went more to it’s company building and to its entrepreneurial aspect.
Before that it was just developing technology and I have to say we have really good mentors at KAT. There’s also schooling program for researchers that want to start a company like really two and a half weeks full time. You learn everything from basics of controlling and finances and financing to design thinking to business modelling and so on. That really helped us to get this off the ground.
Per: I think it’s very important that universities and society in general makes that transition easy because it’s a major commitment to start a company. You have now gone through the first steps of having the company. You have the hardware going from low volume production to more or less serious production. Can you give us an overview of what components you have and why you choose to start with those because I guess in the end you would like to add many more but you have a set now. Why did you choose those? What are they? Why did you choose those and what do you intend people to do with them?
Andreas: Basically we set in order for us having an easier time telling the customer what value we deliver to him. We cannot tell him we give you this but you also need 10 other things to solve your problem. Our main aim was to be able from the onset solve real world production automation tasks that usually means you need some kind of actuation because you are modifying physical products. You might need some kind of sensing and you definitely need to control and HIM modalities of that machine tender can start it, stop it, influence the parameters.
Per: Buttons, screens, being able to tune it.
Andreas: Exactly yes. That was kind of the minimal envelope we require so that we can solve real world tests. That is why we started. Let’s start with the controller. We call it the master module because it goes from 230 volts to our DC bus system of 48 volts. It is also the gateway to the external world so you have a normal Ethernet check there. You have a wifi card or wifi interface. On the other side of the master the robot begins where you have a hybrid cable connection that delivers power to all the other types of modules that connects them by terms of an Ethernet field bus. The master module is kind of where it starts with. There you have the energy. There you have the accessibility from your tablet or your desktop computer. The master module is kind of the control unit can then have any number of the other modules attached to it. Any number of course of limitations in terms of electrical power that is available so with one master you can control up to seven or eight actuator components and of course many more cameras and other components that do not need so much energy. We then continue to the actuators that we got integrated. Linear access because a lot of the systems we found out that are understandable by our customers are more Cartesian systems. Robodev at the moment not at all aiming to build serial manipulators. If you look at our modules it wouldn’t make much sense to build a serial kinematics out of it like a robot arm because the torque to weight ratio is not good enough to chain them together because then the link close to the shoulder or the bottom would just be overloaded.
Per: You will have little payload at the end.
Andreas: Exactly and our goals was to have five kilograms of payload in the end available really for the work piece which means you need more work, you need higher strength for the individual actuators because even if you build a Cartesian system you will have one linear access, the next one connected to it and a cert one that has to be lifted by the very first. We aim to having five kilograms of payload at the manipulator. The linear axis is a very important component. We of course also have rotary modules like a classic for several drive like a drive shaft coming out but also one where you have kind of your aluminium profile based and an aluminium profile as the drive part so that you can easily build a scarab style robot or you can continue connecting the next modules.
Per: Instead of having that axon coming out you still have the same T knot, very smart.
Andreas: Because in the end once you start needing some kind of mechanic adaptor it increases your entry barrier, your effort so much again that we try to avoid that. The actuators as I said integrated linear access, several drives in different variants. Of course the linear access also have different travel lengths from 250 millimeters to 1250 that kind of envelopes manual workplace, manual assembly station. That is kind of our main use case, machine tending and assembly assistance so that is why we do not need an actuator with three meters length but also why half meter wouldn’t suffice. That is the actuation side at the moment. Then we have sensor modules the most important being a smart stereo camera meaning two lenses to optics and an integrated processing system. The images are not transferred as in the ROS paradigm to your central controller and processed there but processed within the camera and only the payload data is transferred over the field bus which gives you much more scalable systems. You can add one camera or five cameras, the controller doesn’t need any more computation of power.
Per: It’s only the result of the analysis that goes out. Your tennis ball is here.
Andreas: Apart from this sensor we will be adding a simpler kind of modalities just the distance sensor and things like that. Those are going to come. We have some helper modules obviously for the vision part. You have a LED light spot in different colors which you can also control over the field bus in terms of brightness. You can blink it, synchronize the camera and so on. This is kind of the sensor part. You have all kinds of I would call them helping modules like you have your HMI panel, multitap screen which you can program for the operator to change the process or to start and stop it. You have an IO module that you can get analog or digital signals in and out and a few other kind of let’s say connectivity modules to existing components.
Per: Instead of having the customer do something beyond your system this makes it easier.
Andreas: We do not aim to get 100% of what is available obviously so in the end we try to provide the actuation and the basic sensing but if you want to do for example screwing applications is a really large demand out there. We see customized ones. We just take a torque controlled screwdriver and guide it to where you need to put the screws instead of developing our own torque controlled screwing module.
Per: It can lead to those. They’re pros at what they are developing.
Andreas: They have developed that since 30, 40 years and it has to be robust in terms of the torques and so on. Certified even if it’s a critical connection. We try to move it where it’s being moved by hand right now. The same is true if you have for example gluing, getting glue out of a nozzle in the right amount, the right speed is really challenging on its own so we will not do that either on our own but take what is out there in the market and integrate it in our system. Just guide it.
Per: But you would use your system to build the overall structure then it would carry special sensors or special tools of course.
Andreas: Yes. There are so many different kinds of tools also for surface treatment and we will not go into all of that. We kind of see as the actuation and sensor provider basically. We control of course the whole process. That is the whole software part I didn’t talk about it. How the Robodev system works is that you put it together aluminium T slip profiles and then you add this hybrid cable. You have the same connection on every device be it actuator, sensor or miscellaneous module and then you turn on power. It is automatically configured so every module has a profile that is being transferred to the controller. Controller knows this is linear x has this certain length, it has this amount of torque available, this position and accuracy and so on. A lot of kind of the meta model of the hardware module so the controller can reason about what it is able to do and configured automatically. Basically for you as a user you turn on the system and then you have what we call the manual control. You see every module that is attached and can by hand change its functionality. You can move linear X as you can turn on the gripper. We got an integrated vacuum gripper as well. We are currently developing also mechanical grippers not like a five finger hand but more what is still standard in industry a two finger gripper and those are of course very important for handlingtasks as well. All these single functions that you can control in manual mode. Z cannot reach the position and so on. See that I can lift up the part. Probably do I need some different vacuum suction cup to test out all of these things. That is manual mode. Automation wouldn’t be much use if you have somebody standing there and controlling the machine by hand obviously. You need to define and automate the process.
Defining the automated process we have a flow based programming approach so graphical interface Simulink by Matlab or (0:24:43.1) or Noderatus I think of a well-known example where you have different nodes and these nodes are interconnected. You start on a blank screen and then you say @component and then you have inputs. You have logical interconnections, you have displays and you of course have your physical modules. You at them to your page and you connect the outputs to inputs of other modules. For example you just have a manual number block that has an output. You have the manual number and your linear axis has an input position, target position then you connect these together. The X has a status output current position less target position reach and a few others and so your process would for example simply look like input of the number block connected to the input of the linear axis target position, the output position reach is connected to what we call a phase change which jumps into the next empty page. Then you could say okay now, move the axis to this position, activate the gripper, the vacuum gripper for example. It has an integrated pressure sensor so you can tell it okay, once you have reached 0.3 millibars of under pressure then the part has been grasped probably then continue to the next step, turn on the light, take a picture, read the QR code and then do this or that. That is how you kind of go through the system. What is really important for us is that the system can be programmed by non-engineers meaning skilled factory workers but not automation specialists.
Per: You don’t have to be a PhD to do this.
Andreas: You do not have to have studied this either. It’s a hands on approach. The whole system is a hands on automation approach in the end. You do not have a CAD model that you build up. You see something needs to move from there to there. There is my parts supply. Here I need to press this part on. The other I need to add glue or insert the screw. You just add it together and how you think you can get to a reliable process and in the same manner you also program the process. You do not have the whole complexity. You can break it down mentally to the individual atomic step. The atomic step for example is move the axis over there, take a picture, read the part position, move in the middle of the park, pick it up.
If you look at a process in the system you usually have less than 10, 15 components on each screen. That is easy to understand. You look at it and you see okay here is this module doing this. This one is doing this. You are not overwhelmed as in texture programming of having a hundred variables having a lot of text there. It’s just easier we think to use it in that manner.
Per: Of course somebody else can then come in and maybe add functionality too for instance the camera. That would also be presented in this more user friendly way.
Andreas: For the camera in particular I think that is also an interesting idea we developed there that we have an assistant like wizard that guides you through figuring out what you actually want to do because assuming I have no idea about computer vision. I just know I want to once there is a part there this process should start. I want to have this binary information parked there or not. Then I do not need a complex set up to define recognizing the part, having a CAD drawing. I basically want to compare more or less two images. Is the part visible or is it not visible? I would get guided and in the end something comes out like for example a template matching approach that will tell me yes, park there or not. Then I will be guided further to actually define the details of what are the corresponding images, what are the relevant regions of interest and so on. If I have the intention as the end user that I want to pick up a part that is positionally flexible laying on a tray without a fixture and I would need to pick that out in the middle then I will be guided through the same process and depending on whether it’s a metal part without surface features or it’s something I can use feature matching I will end up with the right pipeline but I do not need to think about vision pipelines. We rely on more or less engineered pipelines but we make it easy for the user to get to the correct one and to parameterized it.
For him it’s an experience of he gets asked a lot of question, please click into the image, please select the part. Is the right image better than the left in terms of detecting the part? Resource that will only ask the user questions that are easy to answer also for not experts but we utilize these answers for either just an expert system in the background on the one hand and some machine learning on the other hand to get to a tuned pipeline. This all happens in the same web interface. The whole control system is web based in terms of the front end.
Per: I can do that on my laptop.
Andreas: On your laptop or your tablet as if the camera is doing the image processing on its own but you do not need to connect to the camera and configure the camera you just click on the virtual camera on your block and configure and then everything else will happen magically in the background over the field bus.
Per: Which also means that this will open up this kind of automation to a much larger field. It’s flexible so you can do many different things, short order, short runs, you can rebuild your machine but you can also reprogram it without bringing in a PhD or whole team of expensive experts.
Andreas: What I’m always saying when I’m asked what is the unique point Robodev is providing customers. I tell them everything you can do with our system you can do without our system but it will take you much longer. It will be much more expensive. You need somebody external to your company to do it for you like a system integrator but if you use our modules you can give it to your internal tooling department and one day of schooling they will be able to put together such systems on their own and they will not be able to solve very complex tasks initially because they do not have experience but people are creative. They just need to write tools to work with and I think for a skilled labourer, a PLC controller and a servo drive and the CAD program is not the write software to develop automation solutions.
For us we see the big growth of the market not getting more let’s say putting the barrier of what is possible further out there. There were quite big advantages in the last years especially with force controller tests, just having inserting packs into holes. That is enhancing the top of the line. What we are aiming for is bringing what is in principle available but not economical to do to a much larger market by packaging it up into a highly integrated individual components. You do not need to combine your motor with your mechanical axis with your electrical cabinet. You just have this linear axis module. You put it there and that’s it. Then you can start. Taking a lot of complexity out of automation components I think that is different people can use it and if different people can use it becomes possible to use it in different areas because you do not have too much engineers around that are bored and want to build new solutions. They are usually completely have a completely busy schedule. If they can’t do it nobody will do it and it sticks with manual labor. If you have easy to use components you can go into these areas.
Per: What I’m thinking there also is bringing this in house into company and training your staff up to do them. Here, you can be much more nimble. More or less immediately start even manually start producing it and then automation can go on by paralleling and then assume when automation work we’ll do it that way and then people can go on to the next job. This could be part of a workflow for an SME that produces physical things or does physical tasks and really integrating it very closely. The closer to the pinch point, the problem we come, the more nimble we get.
Andreas: Not everybody is a fan of lean methodology but I think it’s really about that that you are close to your production process that you solve problems quick in house so that you actually gain knowledge about your production. You gain know how. It’s also a competitive advantage obviously for you. Just what was lacking kind of was technology that made it easier to do it. There’s this whole field of low cost, intelligent automation where they tried to re-use what is available so it must be cheap, it must be easy to come by. It must be understandable by the people in the production and I think what we are providing is the same with the current stage of technology allows us because a lot of the original books that you find on this topic are from middle of eighties, late seventies even when the whole Toyota production system came over the world. At that time you had your stepper motor. We are not talking about brushless DC motor controls. Those were huge and expensive. Now we can integrate. The carriage of our linear axis contains the motor controller, a lot of sensory processing and all the communication fits into a module like this big. That was not technologically possible 30 years ago. We always run against the ceiling doing it nowadays in terms of integration density but now you suddenly have a module that is really small and easy to handle and you can solve, I think the thinking was already there but you didn’t have the technology to actually follow through on it.
Per: Where are the company now? Can I order these things or they are still in prototype stage?
Andreas: We are just going from having our pre series production in a more stable manner to having a stable series production of the initial 10 module types that I describe earlier.
Per: This is 2018 in the summer that we’re talking about if you view this video which is likely in the end of 2018 and beginning of 2019.
Andreas: Then I hope everything will be available in short order. If you are flexible, if you have either a new order or you just realized this I should change in my production then you do not want to wait many weeks or even months. You can just order them off the shelf and we can deliver it to you within a week or a few days. That is the goal until the end of the year that we have these modules readily available. They are standardized which means they will always be in stock because they are not in variance. The linear axis and you just need to decide on the length. It is not the fastest, it is not the strongest in the market but it kind of envelops the 80% volume that you usually require. It’s usually fast enough. It’s usually powerful enough. The position and accuracy of 0.1mm. Of course you get ones that are even 10 times more accurate but how often do you need it. That is kind of always our goal that these modules provide you for the most common use cases this will be sufficient and this will work. Where does this company stand in another term?
Series production of hardware is starting at the moment. We are 16 employees at the moment which kind of is five times as what how many we were still last year at the same time. Now we are really starting to go into the market. We have many customers waiting to get their hardware shipped because we delivered hardware and systems to some of the early customers but still out of prototype or pre series production just to get even more experience, get feedback, find out if something breaks before you are in larger quantities but that is really I think one of the hard things about producing new hardware. As a start-up you cannot wait to do it probably in terms of build everything, have it test run, have a stress test running for a whole year and then improve it. We do not have runway of four years to do everything so you have to kind of interweave this. Have early prototypes run them in house until they break. Figure out what broke first. Improve that and go through a few of these cycles. Then go with a few customers that know that must be honest to them and tell them this could still break but we will obviously give you a replacement module within one or two days.
Per: Because you don’t have 10,000 different variants. You can do that. Even as a start-up you can.
Andreas: That we can. While you are doing that get your series production running, start with a little bit smaller batches to produce. We are talking B to B industrial components. We will start with like on the order of 10 to 50 modules per module type and not hundreds or thousands. We hope to get into that company soon obviously but that is kind of how you can do it at all.
Per: You’ve started to get these systems out there, these prototypes. You’ve also used them yourself for many tests. Can you tell us a bit about some cool things that you have done with them and how the process of getting those things done differ from doing them with the old school stuff?
Andreas: There is a good and bad thing about having not enough hardware around. The bad thing is you don’t have enough hardware around for development, for stress testing and so on. The good thing is you are forced to repurpose it nearly on a daily basis. What I brought with me here for our interview I took out of our trade fair exhibit that I used last week. Next week it will likely go into a showcase for a customer again. We are constantly taking these apart and putting them back together. Everything that will annoy a customer later on because it’s hard to put it together.
Per: Annoys you now.
Andreas: Annoys me so much so we will definitely change it before it annoys the customer.
Per: It’s called dogfooding in coding I think. If you don’t use your own code it’s never going to be worth anything because scratching your own itch is so much likely to happen than scratching somebody else’s.
Andreas: What we really figured out soon is that you can build also some fun stuff with it. It’s called the founder barbecue which in Germany (0:41:33.4) is kind of a play of words with barbecuing the founders and the founder is providing a barbecue.
Per: We’re going to roast you.
Andreas: For that we built a small automated grill, a small grill robot but really simple the (0:41:48.3) arm I think it had two and a half (0:41:52.7) but we really put it together in an afternoon. Then we cleaned the modules again and showed them to customers the next day for a totally different setup for machine tending. Being able to do that you cannot do that with what is out there as in terms of automation components nowadays. That kind of really gave us the feedback okay, we achieved much better usability, much better reusability in terms of putting together physically and also programming it because your process looks different for turning around sausages on a grill then to putting parts into a machine. Also I’m amazed it looks that not much different because in the end you’re moving to certain positions. You have certain sensor triggers to continue. You can actually recycle some of the software process between those two as well.
Per: I don’t have personal units of this magnitude but I have lots of kids systems all the way from small kindergarten all the way up to kind of high school anyway. I did so much fun to build stuff and build stuff. I clearly see this that you can use it as a barbecue robot one day, take it apart and then use it to show a customer. The only problem with that is the customer is going to can we go to lunch. I’m hungry. I don’t know it. For some reason I remember a hamburger. Nobody will ever know why but I really believe that this is the power of the systems. When you do this it improves the product on a fundamental level but as we say bringing this in house and doing this for a particular business it won’t only automate a particular task it will allow you flexibly and quickly automate any task within your business but it will also tell you a lot about your business because when you automate it you’re forced to think about the way you do things because usually you have to change them a little bit when you flip the burgers that you didn’t do that a human would do it in a way that the robot could. We adapt the world to the robot and the robot to the world rather than just one of them because we can’t make a robot that it’s unnecessary hard to make a robot that could it exactly like a human.
Andreas: That was actually one of our motivations to go with a cyber-physical system because we did a lot of programming robots during our PhD and we realized it’s an interesting intellectual challenge to get serio kinematics to do some things with a gripper and really sensitively pick something up and put it in somewhere but from a holistic point of view you would not do that. You would just have a mechanical structure that eases it to do the task. You adapt kind of both the process and the machine and when you want to adapt the machine it means that it has to be modular from my point of view. I have no clue how I would get a machine that is easily adaptable to what is good to the process instead of just having a fixed monolithic machine if the machine is not itself a building kit or a modular system.
Per: This modularity is in many respects very hard to achieve. We talked about self-reconfiguring modular robotics which is ridiculously hard but when you solved it you have solved a larger problem. When we talked about 3D printing we sometimes say that complexity is free. When we do the regular C and C milling and lay the work complexity is very expensive. Every move, every time you modify the work piece it has a work piece associated with it. When you 3D print, 3D printing a square and 3D printing a flower petal more or less takes the same time. If the process can do it complexity is free. This is very smart. This is an illustration of the true power of modular robotics is that we move complexity up towards what you’re doing. You are doing the complex work and the hardware and then we remove it from the later stages. Since we only have to develop Robodev once but we can use them in any number of situations than total amount of complexity is vastly reduced.
Andreas: Also the cost associated with it. We have standardized modules that are produced in a much larger scale. It gets much cheaper than if you would develop because that is kind of what automation is doing nowadays. They are building what we are putting together as standardized components often from scratch each time. They’re integrating motors with their drivers with sensory data and if you do that once in a more or less general purpose manner you can get economy of scales gains. What is also I think important about the whole aspect of having systems at the customer so we will do both. We will focus on certain applications and provide you with a blueprint. If you want to put together parts by means of screws perhaps you should build it kind of like this and the kind is an important, is the flexibility, if you need motion over large distances you just use a different module and so on and so forth. If you need to integrate it somewhere special you might change it around a bit but focus for applications. What I think is the more interesting part is companies having these modules lying around because they are either disassembled an old application and they didn’t re-use them yet and it lying around and then they will get ideas. Couldn’t we do X?
I already saw one of our customers doing something that I would never had the idea that you could do that with our modules. We will see much more of that because we do not have too much systems at the customers at the moment. It’s already starting that our customers figure much more out what you can do with our modules than we know on our own because we are not experts of the customer production. The customer is hopefully the expert of the customer production. That is kind of the same if you look at what people have done with excel for example. It’s amazing what you can do with a spreadsheet calculation. I’m pretty sure whoever the programmers of the spreadsheet application have no idea, I think somebody reinvented kind of version control in some system and a whole ERP system. You can do that. It might not be ideal but if it solves your task and you can quickly get to the point with it then why not. That is what I think is really exciting. We hang up a large banner in our first basement office of a research institute. I think it’s more common in Germany than the garage at home and enabling people to make things work. That is kind of what we are really aiming for just handing out these modules, these building blocks to people and then they will figure out amazing things to do with them but they could not do it if we wouldn’t hand them these tools.
Per: That is what I’m really excited to see in a couple of years. What I think there is I always have that modular roboticist crack for engineers. I can tell you who it is. When you have them there it happened many times that I would rather build something than do what I really should. I think that is a message to companies out there if you think this could be relevant for you, acquire a number of modules and you just put them in the hands of your people because they are I’m sure they create things with them and even kind of if they don’t they will learn about your production and they would be able to improve your production just by the fact that they think in new ways and considering how you do things and why you do things that way and stuff.
Andreas: For prototyping obviously that is also a good approach. You use a Robodev system to figure out and not just on paper and not just in cardboard but in real actuated hardware how a special purpose solution could look like and then you might still want to build a special purpose solution if you need much higher throughput for example. In the end if you know you will run it for three plus years it pays off to build it out of individual components because then you can guarantee a higher life cycle and a longer lift of the special solution but getting there from just from paper usually results in systems that won’t work the first time around because you forgot something. You didn’t consider a certain aspect. If you prototype it in reality. In reality you cannot forget that you cannot reach the part. In paper it’s easy to forget such things.
Per: We also know this from simulation that everything is caused with the success in simulation. It always works in simulate because we can’t simulate the world in detail. What I also think is when we think about this custom built machine that is more efficient in three ways but it’s also going to break down and it might take weeks or a longer time to get a spare part an in business it’s a big decision to try to look three years into the future. We might have another financial crisis. A competitor might move in or whatever is going to happen. I think using modular robotics in this way is kind of an insurance policy. Being off the shelf it’s just going to come to you in basically hours, overnight shipping in Europe and US is very efficient. Also you will be sleeping better fi there is a business change. We haven’t sunk all our cost into this.
Andreas: It definitely decreases your investment risk because if you invest in automation nowadays you usually invest in the belief that this particular task would be performed over this amount of time and with a higher volatility in the markets as you see nowadays and I’m pretty sure this will grow instead of going back. You just cannot guarantee the time anymore and perhaps your product has changed slightly to either to follow a competitor or bring something new or differentiate yourself from the competitor. Then your fixed automation solution doesn’t fit 100% anymore and your investment has been sunk but if you can retool it in the worst case you can just take it apart and repurpose it later on. You’re safeguarded against having sunk a lot of money into a solution that you will not use until it has paid off. You can I think take that even further why even buy it in the first place. Why not just rent the individual components because if you look at what is being rented out nowadays and what is not it’s really telling. It think nobody at the moment is renting robot manipulators in larger quantities. Also it’s a universal motion machine but you need to add too much infrastructure around it usually. You have to integrate it somewhere. They are not being rented. C and C mills for example they are being rented on a daily basis if we talk about industrial production now because it’s a standardized machine with a standardized task. You can just pick it up with a forklift, put it somewhere else, put it down and start using it again. You do not see that a lot in the area of automation or robotics. If you can get there that your automation technology for assembly, for special purpose solution becomes rentable.
Per: When we add the financial flexibility of renting short term, long term or for a specific time which of course brings that cost down or for an ad hoc time this will, research and development would be much quicker. Prototyping will be very much cheaper and easier but also this again addresses the business risk. You will be able to go into many more projects and explore them because you know that you’re not tying down a huge amount of money for three years. Say that you run a business and you have a certain capacity to explore different cooperations or take in business. You will be able to greatly widen your ability to take in business. We don’t know what becomes to success. Many products are brought to the market. Some are widely successful. Some are less successful and some are failures. If you have a wider net you are more likely to catch the successes. If you can have a wider net with less risk that of course also on the macro level your company becomes much more stable.
Andreas: In an Austrian production magazine there was recently an article about what they call about instant factories or pop up factories if you look at this concept of a pop up store. There is a brief customer spike demand and you want to fulfil this like in the initial have a start-up but it will not be profitable in a year.
Per: Not even all of the year. It can be seasonal. Not even selling flip-flops in the summer and winter caps in the winter.
Andreas: If you have these large swings in demand also your factory, your supply has to adjust to them and the message in this article was in the 21st century of course speed is king but not just in production as well but not just speed in terms of throughput but in terms of how fast can I put together a factory to produce the part that is currently highly on demand. Let us take the featured spinner. I read the statistics correctly if you miss the start of production by two weeks you miss more than half of the market. Usually people that I’m talking to that have large factory automation equipment put into production they are talking about months just taking it into production not even planning not considered. It’s half a year or year more and getting to technology that can deal with this kind of rapidly changing demand will be different technology than we see nowadays. Whatever is out there nowadays is still growing and will be let’s say the stable products. We will always need toilet paper and toothbrush and so on but for products that are kind of seasonal or that are hyped or for some reason for that you also have to see different technology. That will pull all through the supply chain because obviously somebody will assemble the final part or the final consumer good but his suppliers will then also have to react quickly. Since everybody has to stay lean and not fill huge warehouses with goods that won’t be demanded by the market anymore half a year later and then have to be written off you have to production on demand. This on demand through the supply chain when your demand is highly volatile that is really the interesting challenge I think in production nowadays.
Per: I know there’s lots of production in Europe here in factory for a day.
Andreas: It’s a great project.
Per: I’m always talking about the factory for a day so that we set up one day, we use it for one day. They break down and we set up a new factory. You will have elasticity to your capacity. You can add very quickly. You can reduce very quickly. I think this is truly the way forward. Modular robotics are very suitable for this model both for the provider and the consumer. This of course makes it a good deal for everyone.
Andreas: That is kind of the same. I always compare it a little bit with the computer industry nowadays you can get so many components that are interchangeable, your server, your desktops are also modular systems. If you need a lot of storage you just get your standard PC but add more hard drives. If you need a lot of computational power to train your neural networks you add a few GPUs and only one hard drive and so on. They’re having the same flexibility will also pay out as industrial automation market because it just does not make sense to have custom tailored production sales on storage for rental because there will be exactly one customer that ever request it. Request one of those and your warehouse is full and it pay off as a rental model. If it’s just components of automation then it looks totally different.
Per: It’s been fantastic to talk to you about this. Thank you very much for taking the time to do an interview. It’s been a pleasure speaking to you.
Andreas: Thank you very much for having me.
Per: Thank you very much. I hope you liked this episode of the podcast version of Robots in Depth. This episode is produced together with Wevolver. Wevolver is a platform and community providing engineers informative content to help them innovate. It is how engineers stay cutting edge. Aptomica is the founding sponsor for Robots in Depth. Aptomica runs anything in modular robotics. Dream, rent, build.
END OF TRANSCRIPT