Mouser Talks Tech with onsemi about key technologies powering Autonomous Mobile Robots | Mouser

Added: 2026-05-28

[00:00:16]Stuart: Hello and welcome to this Deep Tech  interview with OnSemi for Mouser. I'm your host, Stuart Cording, the Electronics Reporter. If  I were to mention robots in the context of industrial automation, you're probably thinking  of robot arms welding car chassis and installing windscreens. In these examples, the robot arm is  bolted to the ground and fenced off for safety while the workpiece is brought to the robot.  But in today's world of increasing efficiency and optimization of menial tasks, we're seeing a  growth in untethered robots that operate alongside humans and move to wherever the workpiece is  located. Although such autonomous mobile robots, known as AMRs, vary significantly in design,  they share many common characteristics.

[00:01:03]These include their own power source, a  wide array of sensors, connectivity and increasingly some level of intelligence to  tackle the unexpected autonomously. To get the lowdown on the latest technologies that are  making AMRs possible, I've invited Theo Kersjes, Global Application Manager at OnSemi, and Aly  Barakat, Technical Marketing Engineer at Mouser, to share their insights. Theo: Hello, Stuart.

[00:01:29]Aly: Hello, a pleasure to be here. Stuart: Well, it's great to have you both. And Theo, I'd like to start by asking  you the first question, which is to expand on that definition of AMRs I've just given,  perhaps sharing what you and your applications team are seeing at the customers you support. Theo: Yeah, currently today it's a very broad group of different types of robots and I have a  slide with some definition and characterization of each of those. We might use that quickly  to go over the slew of different robots. And as you mentioned, on the left side on this slide,  there is the autonomous of the big robot that was used in industrial applications, which was fenced  off. Next to that, there was the automated guided vehicle, the AGV. This had a line on the ground  and it could follow that line, and it would stop if there's an obstacle, but it cannot go around  the obstacle. So these AGVs are very clear to people if you're working in the same space, what  they're doing and how they're behaving. And that's one characteristic that we need to keep in mind  when we go to the newer systems. Secondly, or the thirdly, then the autonomous mobile robot you see  here, the warehouse robot that got that name that we're now using for this whole category, where  Amazon currently operates more than a million of these robots in their warehouses. And these  first three initially were all programmed with traditional software methodologies. And slowly we  got some machine learning, AI and other techniques in there that were required to even control more  motors more dynamically. And that's why the cobot, which is the next category, got its popularity,  the six motor arms. Now it's getting hard to control all these motors to go to a certain  physical position in space. And that also uses waypoints. So you can set waypoints by moving the  end effector or the thing that's attached to this arm to do something to different locations. And  then the robot will automatically fill in the path while considering safety, keep out areas and other  things. And that's where I met Aly during the cobot activities we did at OnSemi. We have a lot  of motor drive, motor control for these systems, power solutions, AC-DC compute. And we had a demo  at a trade show and Aly came over and he said, that's cool. I wanted to look at that as well. And  so that's where we started our cooperation about, I think it was three or four years ago. Stuart: Wow.

[00:04:32]Theo: Yes! Stuart: That's how it starts, isn't it? And I know Aly as well. And I can understand  his interest in such an application. I would have done exactly the same if I'd been in his position.  Now, Theo, would you say then that there's sort of a standard architecture for AMR's emerging,  sort of comparable to the standardization we've seen in PCs? Or do AMR designers  tend to sort of optimize the architecture to the task and the deployment location? Theo: Yeah, so definitely there is commonalities.

[00:05:07]So after this cobot, now you see a standardized  platform emerge. You see an arm on a mobile base, which we call an autonomous mobile manipulator.  And then from that, then the humanoids come as well. And these all are battery operated.  They all have battery charging capabilities. Then they need certain sensors to be able to do their  pod planning and driving safely around people, because we now want to use these more flexibly  outside of their warehouse, maybe even as a delivery robot in our own neighborhoods or on  agricultural. Or agricultural settings in fields where sunlight might be effect or shadows, hard  shadows, where sensors can be overloaded. So yeah, there's a commonality between them. If you split  them up in power levels, you can even say there's a lot of commonality with the power solutions as  well. Of course, if you go higher power solutions, you come into the EV type of silicon carbide  solutions for traction inverters, which on Semic also has. But with these robots, it tends  to stay below four kilowatt, let's say that. So a golf cart size apparatus that does something. Stuart: Fantastic. Now, one of the things I mentioned at the beginning is that those  industrial robots that we've typically seen in sort of marketing films, they're bolted  to the floor, they're in a safety cage. And the whole purpose of that is also to avoid injury to  humans operating in the vicinity. But these AMRs, not only do they need to sense their  environment to ensure safety, they also need some sort of spatial awareness as well. Are  designers relying purely on one type of sensor, like a camera comes to mind, or are they combining  the inputs from multiple sensors to do that?

[00:07:03]Theo: Yeah. Some designers have a whole philosophy  around this and like to use one type of cameras, or in most cases, a camera sensor. But that's not  really cost efficient. The compute needed for a spatial image segmentation or something you want  to do in an image camera is way more than using, for instance, an ultrasonic sensor. And other  techniques like an ultrasonic sensor are very good at detecting glass objects or objects that  are hard to detect with an image sensor. So if you would have a full sheet of glass or some  other obstacle, mirrors, you can think of if you have robots that operate in spaces where  humans are, you have these fancy mirrors in spherical shapes to look around the corner. And  a drone with only an image camera might have a hard time and thinks it can fly out the mirror or  doesn't know that it's around the corner. Whereas the ultrasonic sensor immediately detects this  is a solid object. It must be something else.

[00:08:19]So definitely ultrasonic, but also LiDAR and  other technologies, other sensor modalities.

[00:08:27]Stuart: Now, I've taken a look at various  image sensors over the years and I will be perfectly honest and say I'm still a bit  unclear on what impact features like the rolling shutter and global shutter have. So  what are the requirements AMR developers have of image sensors and how do they then go on to  determine which optics match with that sensor?

[00:08:47]Theo: Yeah, definitely for spatial awareness,  you need some kind of depth sensing and OnSemi just launched our latest ITOF, indirect time of  flight depth sensor, which is an image sensor, 1.2 megapixel high resolution that from the  camera immediately provides a depth field that can be used to make this spatial awareness  in some navigation applications. In general, you cannot solve everything with one type of image  sensor to create one type of camera. And you can think of some of the pixels that you create in  the image sensor might be more light sensitive to light or to color, and that might be helpful in  certain situations. And so we created our Hyperlux series of different cameras. Some are very fast.  And so if you have a stereoscopic application where your robot looks into its surroundings and  wants to do a depth field with stereo cameras, you need very fast responding cameras because of  the fast moving objects around the robot. In other cases, you can use the high resolution cameras for  detecting some specific application it's doing. So we have these categories and these categories are  then identified by some key characteristics like, for instance, high dynamic range where you might  think when you're driving a tunnel and you all of a sudden come outside, the light might blind  you for a second. So we have cameras that can fix that and just look out of these scenarios  and still see clearly what's happening so the robot can determine what to do. And this is also  what you need with industrial robots when they go into shadowy areas or come into agricultural  field applications with stuff like that.

[00:10:53]Stuart: That's great. That's one of the big  challenges that you see is even with the phone cameras, for example, that challenge with big  differences in light. And it's incredible to see how that's been solved at the semiconductor level.  Now, power is obviously another key challenge in AMRs to get the most out of the batteries, first  of all. And then you've also got the charging and the power conversion to an efficient motor  drive. What demands are being placed on your team with regard to that part of the application? Theo: Interestingly, the owners of these robots have the same problems as EV owners. They had  some range anxiety on the batteries. You know, is my robot be able to complete its whole mission  during the shift time or the runtime I want to have out of it? And there's different ways of  solving that. Of course, higher efficiency of all the components is always a drive to that.  And I'll come to that in a minute further.

[00:11:55]But another thing for warehouse robots where you  have many of one type of robot, you can add some robots if it doesn't fit your profile and then  charge them. You can do intermittent charging or opportunity charging where they quickly charge  if they have nothing to do between jobs. And we see already some humanoid robots that exchange  their own battery packs, which looks really nice.

[00:12:25]But that implies on a chip level that you have  something like a hot swap or e-fuses where you can protect that process and handle it. And there  again, we have many products that can assist with these kind of applications. Lastly, I mentioned I  would say something about performance. So motors are in humanoids. There's about 30 motors. So  they're one of the most complex systems that are also very different in humanoids. And for each  of those, you might have slightly different power solutions to control them. And it's all BLDC most  of the time. But you might want to have a higher efficiency to reduce the heat and to reduce  the demand for cooling aluminum or cooling plates so you can make this system lighter. Stuart: Super. Well, that's a great overview of the basics of the AMR application and all the  elements that come together to make them. Aly, I'm going to come to you now. Now, for those  who had the opportunity to visit Embedded World in Germany this year, they'll recall quite a  large AMR robot roaming the Mauser stand. Now, that was constructed by you and your team. Can  you tell us a little bit more about the project, what its aims were and the AMR that resulted? Aly: Yeah, absolutely. So the idea behind the AMR project was really to show what the next  generation of factory automation might look like. So where robots aren't just following  fixed routes, but actually understand and adapt to their surroundings. So together with Theo  and the whole OnSemi team, and as well as Mouser continues expanding its industrial automation  portfolio, we wanted to build something that connects the components we distribute to real  working applications. So we developed this large human safe AMR equipped with multiple sensors such  as 2D LiDAR for mapping and obstacle detection, stereo vision for depth and odometry, mono cameras  for object recognition, and lastly ultrasonic sensors for close range safety. And then this all  supported by encoders and an IMU. So at Embedded World, we set up a live demo where a cobot placed  a box on the AMR, which then navigated through obstacles to a precise drop-off point. And the  goal really here was to highlight how intelligent, sensor-rich AMRs can enable safer and more  flexible internal logistics in warehouses and factories, and to show as well that Mouser's  technology portfolio is already powering the shifts toward smart autonomous factories. Stuart: Now, as mentioned by Theo earlier, AMRs can't solely rely on a single type  of sensor or vision system. So instead they use a system called sensor fusion, which  combines input from multiple sensors to improve decision making over a potential single sensor  approach. Can you tell us a bit more about how sensor fusion works and how does it figure  out which sensor is the most reliable to make the decision that needs to be made next? Aly: Well, that's actually a great question and that's really at the heart of autonomous  mobility in general. So AMRs essentially works on probabilities, so nothing is ever perfectly known.  So the most important things are perception and localization. The robot then constantly needs  to estimate where it is and what's around it, even with noisy or incomplete data. And  that's where then sensor fusion comes in.

[00:16:03]It combines them all those different inputs to  build one consistent, more reliable understanding of the world it's in. So now there's a lot of like  sensor fusion techniques, but the most like common use technique in autonomous mobile robotics is  based on Kalman filtering. So essentially the robot predicts its motion from encoders and IMU  and then it corrects that prediction using data, for example, from the lead, our camera ultrasonics  and so on. So the filter's job is like basically to automatically decide which sensor to trust more  depending on the situation it's in. So let's say, for example, the visual odometry becomes  unreliable due to motion blur or lighting changes in a warehouse or factory. Here then the  LiDAR comes and takes the lead. So then in the AMR that we built, we use an extended Kalman filter,  which literally like handles the nonlinear typical behavior that mobile robotics face. So LiDAR  gives precise geometry. Vision then adds like depth and some context, like object recognition.  Encoders keep short term tracking stable and ultrasonics lastly ensure like cross range safety.  So by fusing all of these together, the robot then reduces uncertainty, localizes more accurately and  can navigate then smoothly in dynamic environments where humans are in. And that's really then  what makes the sensor fusion like the backbone of any modern autonomous mobile robotics. Stuart: Fantastic. Thanks for that insight into sensor fusion. This made it a lot clearer  for me. Finally, one last question for you, Theo. We've covered and touched on some  of the silicon devices which are used and required to make a functional and reliable  AMR that will operate as long as required in a factory or in a logistics center.  What else is needed and what requirements are impacting the choices made by engineers? Theo: Okay, yeah, we talk now about the chat GPT moment for robots. So we find in their tasks  and behaviors that they can be applied to many different applications that opens up different use  cases. And for that, of course, we need the better algorithms. So there's a lot of happening in the  algorithm space and in the way we train these robots, so simulation environments and accelerated  training in in the simulation space. And this is currently focused a lot on the behavior or  functions of the robot, but you can bring this down to even the chip level where you look at  performance of our battery, how long the battery lasts. And so that's for us the angle of using at  the simulation spaces. So that's the algorithm.

[00:19:03]The next thing you need is better mechanical  solutions. And there's some very interesting gearboxes for motors. One is the Ashimedes Drive,  which is basically gears that have no teeth. And they work like a train where the train wheels are  smooth on the train tracks. And so you don't have any backlash in these solutions. So there's very  different advancements made in the algorithms and in the mechanical solutions. And then, of course,  you need different chip integration solutions.

[00:19:45]And for that, OnSemi announced about a year ago,  we have the trio hardware development platform, which is our foundation for next generation  power management and sensor interfaces as well as communication chips. And this trio platform  is a bipolar CMOS, DMOs, 65 nanometer technology where we can very quickly use proven IP  blocks to create different functionality in the chips that is needed. And this then  supports fast time to market, which is then needed for these robotic systems where proven  functions that are in an algorithm can be placed in the hardware and used in our products. Stuart: Great. Well, thanks. Sounds like an exciting next step. And I think it reflects as  well the way the semiconductor industry sort of integrates further as applications take shape.  And we understand the challenges involved in a particular domain in more detail. So thank you  for that. Now, as we always do on these little videos that we record for Mouser, we reach out  to our community to ask them if they've got any questions on these topics. And luckily, I've got  a few notes here. Now, the first one actually came up while I was talking to some people at Embedded  World, funnily enough. And the question was about how this project started. Theo, you mentioned  already at the beginning a little bit about what happened and how Aly took an interest in the  first demonstrator that you had. Can you give us a bit more background as to what happened  next and how you've been working together?

[00:21:27]Theo: Yeah, so one of the first applications  OnSemi worked on was the fruit sorting algorithm using our image sensors to detect if fruit  is fresh and then bin it on a conveyor belt using a robotic arm. And that's what we  demonstrated. Aly saw that and he said, you know, can we do something where we would have  a water dispensary system where it gives a drink to people? And we created that and it was a lot  of fun. There was one of the first with a little time crunch at the trade show doing some waypoint  adjustments. And then also some integration with the water dispensary unit, which had an inductive  button instead of physical buttons. So we had to put something on the robotic finger to make that  work. And then they just took it, Aly took it and they made several different things like ice. I  remember it gave soft ice. They were giving that out and adding conveyor belts and just put their  whole own spin on it. And we went further on the mobile robot then. So, yeah, so that's where we  engaged. And he's adding challenges to add more sensors and look at different conditions for the  sensors in the system. And you could think of like automotive. We have ultrasonic sensors that are  in under the car and in robotics. They can be used for that as well to see if there's anything under  the robot before it drives away. So many different things that we look at. So that's really nice. Stuart: And Aly, both you and me visit regularly different trade fairs across the country and  we see lots of different demos. Some of them are most incredible. But what was it about  the AMR demo that you saw from OnSemi that made you think, yeah, I need to talk  to Theo about that. That's something that we could build upon and do more with. Aly: Yeah, well, first of all, the collaboration, as Theo mentioned, how we first met and how we  collaborate together with this water dispenser robotic arm demo. We had a really good  connection from the beginning. So like we we enjoyed the time together building it and  staying as well after hours in the trade show, trying to figure out everything and make it work  seamlessly. So, yeah, we just visited their booth and we saw their AMRs. And first of all, it really  looked amazing aesthetically. It looked really, really nice. The car was built fantastically.  And I said, OK, let's build this. We want to build this as well to have at the Mouser  office. So that's where it's kicked off.

[00:24:30]We began then building one together, making  it more advanced, adding more sensors to have it more reliable. And yeah, we then said, OK,  this will be our biggest collaboration because it was really interesting to build the car from  scratch, 3D printing everything, making the wiring harness from scratch. Everything is like handmade  and so on. So it was a really good experience for like our Mouser team as well to begin diving into  this world of autonomous mobile robotics and as well as we're transitioning more and having more  of the industrial automation components in our stock that we have hands on experience with them. Stuart: So, Theo, another question that's come up here on my community questions is about AI  and edge AI. How are you seeing developers integrating this into their products? Theo: Yeah. So today you see I mentioned already physical AI. That's the next level after  agentic AI to have robot spatial awareness in the physical world. So then when a robot does a task  at a table and something drops off the table, it will know it's out of its visual range. It cannot  see it anymore. But if it would drive around, it might encounter it on the ground. So those kind of  aspects where we naturally think about as humans, that's not something that's getting into these  models for these robots. And they call that physical AI. When they learn that, then you have,  again, a lot of different use cases that you can add and service with robots. And for that as well,  we have some hearing aids products and we see now some use cases where the hearing aids can be put  on the robot in its head, where you have kind of the center of the sensors. Where you have your  vision system, your hearing system. And then the robot can use this audio to do noise cancellation  in very loud environments as well as directional hearing. If it looks to you or it can detect that  the speech coming from a certain person and do something with that. And so that is the way we are  able to instruct the robot what to do with audio, with text. So that's the prompt engineering of  the robot, as you say. So we don't have to type anymore. We can just talk to them and engage with  them very as we would with other people around us. And it would give again another level of use  cases and enablement for these type of systems.

[00:27:24]Stuart: Wow, this is really exciting and it's  interesting to speak to engineers who are out working closely with customers to understand  what's being discussed at that cutting edge of decision making around new products. I've  got another question here on my list for you Theo as well. So what are the most unusual  use cases for AMRs that you've seen so far?

[00:27:51]Theo: Yeah, for me it tend to be around solving  some really clever problems. And one of the things I saw was a robot that sits on the hull of a  ship. It's magnetically attached and it would clean the hull of the ship like almost your vacuum  cleaner at home, the little Roomba thing driving around on the hull of the ship. But it could do  that even during travel. So for going into port, it could do that. And they were even now looking  at sanding that paint off so they could have the ship very short time in dock to repaint. So  time it just right that it arrives with the robot being done on the hull and do these kinds  of applications. So that was one. The second one is of course humanoid robots. I was two weeks  ago in a large trade show on robotics and it was really cool to see before we see now every week  these humanoid robots do fancy dance. But there they actually had six or seven of them sitting on  the conveyor belt and working together creating some products. So that was the first time I saw  that. And with that also the hands of the robots, or we call it the end effector. So they're more  human like hands and all the new things they can do with these hands is also very interesting.  This is really cutting edge. Until now, robots are doing pick and place. So you order something at  Amazon or somewhere. They can go in the warehouse, get the product, put it in a box and ship  it. That was the most complicated task for them. And now it's doing anything where there's  abrasive behavior. So polishing or sanding. And that's very challenging. And people look at use  cases to do that for robot hands so that those are some of the advancements that are going on. Stuart: Amazing. And I saw another video from Scheffler, which was a similar humanoid robot  where they were working alongside humans, but to carry crates from one location to conveyor  belt. And it does very much seem like the humanoid robot is going to be an increasingly important  part of factory automation. Now I've got one more question from the community that came up. And  I think this is probably for you, Aly. So the question is about operating systems. Now we've  got operating systems which are quite complex, like Windows and Linux, and right through to very  basic stuff that runs bare metal at FreeRTOS. But the question is that they've heard about a robot  operating system, which is known as ROS. Do you know any more about that and how it compares  to the operating systems that many of us know?

[00:30:47]Aly: Yeah. So the robot operating system, in short  'ROS', makes robotics much more easier in general, like robotics implementation, because there  you have packages, specific packages that you can just use and then combine them together to  have your specific application met. So ROS here operates in nodes and topics, like in a system  called nodes and topics. And what happens is each node causes specific topic and this topic  causes another node and so on. So in the end, you have like a workflow of what you want to do  with action, states, input, outputs and so on.

[00:31:19]So in the end, it makes everything much easier.  Now ROS here is based on Python and C++. So you have parts where you can program it using Python  via Python and parts you can program via C++. So it makes it as well much more easier for people  to either program it in this way or that way.

[00:31:39]Stuart: That's great. Well, thanks ever so much.  That's the end of our community questions and also the interview. Thanks, Theo, Aly, so much for  your time and sharing your insights with us.

[00:31:50]Theo: Thank you, Stuart. Thank you, Aly. Aly: Thank you so much. It was a pleasure. Thank you, Theo. See you soon. Stuart: Well, that's all we have time for.

[00:31:59]So let's review what we've discussed. Robots are  no longer bolted to the floor inside safety cages.

[00:32:06]Instead, they are increasingly mobile, interacting  closely with humans and are becoming intelligent too, capable of handling non-uniform loads and  adapting to the environment around them. As an engineering task, the development of AMRs is  exceptionally exciting, requiring expertise across a broad range of electronic systems.  This ranges from power distribution and motor drive systems to wired communication, battery  charging, lighting processing and sensor fusion.

[00:32:35]OnSemi can be found supporting each of these  challenges with products such as the highly integrated mix signal trail platform and their  Hyperlux image sensors and time of flight sensors.

[00:32:47]The recent joint project between OnSemi  and Mouser on their AMR demonstrator also highlights another factor critical to a project  success, collaboration. Only when engineers work together is it possible to overcome the technical  challenges that arise during the development of such a complex system. My thanks go to Theo  Kersjes of OnSemi and Aly Barakat from Mouser for sharing their technical expertise around  the topic of autonomous mobile robots. If you're keen to learn about power efficient motor  control, lighting or the highly integrated sensing technologies that are used in these applications,  follow the link to the Mouser website. [Music]