Inside the Heart of Robotic Motion - We embed TI in the robotics podcast

00:00:04: Robotik in der Industrie.

00:00:06: Der Podcast mit Helmut Schmidt und Robert Weber.

00:00:08: Hello everybody and welcome to a new episode of our robotics podcast.

00:00:15: My name is Worldweaver, And I'm sitting again in Freising next to Christian and Thomas.

00:00:21: We want talk about the whole topic of motor control sensing and power stage

00:00:41: for industrial systems, focusing on robotic applications.

00:00:45: Working for TI now for twenty two years?

00:00:48: Yeah thank you

00:00:49: and Kristen.

00:00:50: yeah my name is Kristin Mogensen.

00:00:52: I'm working in the same team as Thomas focusing on robotics here also.

00:00:56: i've been working at TI four eighteen years.

00:00:59: so we have some interesting topics today motor control it's one topic sensing is another and power stage.

00:01:06: how are these three play together in the robot, Christian.

00:01:12: Okay so when you look at the robot right?

00:01:14: You have to think of... The movement of the robot is always based on the motor.

00:01:24: typically gives you the best torque performance.

00:01:33: So, that torque you need to actually do the actual movement and you can move enough weight right?

00:01:38: And then you need the electronics to actually move the motor which is more interesting part for TI of course!

00:01:43: Now with that piece You have a small subsystem.

00:01:47: you need implement That small subs system... ...you need to have real-time communication through robot controller.

00:01:53: so we talk about it.. ..you also needs a real time processor which has specific hardware features or hardware sync features in the peripherals that you can measure and control, the motor control.

00:02:07: In the right time?

00:02:08: In The Right Windows?

00:02:09: In the Right Time Sync.

00:02:11: because You Can Imagine if you let's say start controlling the motor but the angle sensor is not in sync with The signal you give to the motor in the right way, You're basically introducing losses to this system because it's inefficient.

00:02:24: Because your always trying to regulate something that is not what real status.

00:02:28: and This we have specialized with with TI processors.

00:02:32: now for last I would say twenty five years or longer With first C- two thousand family but also now they are AM Two Six Family And also the AM Two Four Family That We Have Very Tight Peripheral Connections For Right Features To Do a very time sensitive motor control in a very deterministic way.

00:02:52: Thomas, you want to add something when it comes to sensing?

00:02:54: So how do you combine those three topics then?

00:02:57: Yeah,

00:02:58: actually for the precise movement of the robotic arm or also for a mobile robot.

00:03:03: Or for the locomotion of humanoid robots these Three systems need to play together in real time.

00:03:10: so The control loop needs that information from the sensing portion and real-time To control their currents for the motor.

00:03:18: And here our real-Time microcontrollers Actually with the corresponding interfaces through the encoders between the different motors are crucial to achieve those real-time communication.

00:03:32: Do we underestimate these topics a bit when it comes?

00:03:37: I

00:03:47: think it is a little bit ignored.

00:03:50: A lot of it when you go in the whole discussion with this simulation models, right?

00:03:54: You simulate that the motor's moving at specific way and a specific speed without specific dynamic performance And then you just assume This Motor Drive can do It!

00:04:05: And this Is something where actually need to verify.

00:04:07: this Can be done... ...and second piece which is an even more crucial pieces.

00:04:11: The motor control gets hot.. ..and you can Think Of If Take a Humanoid the surface or any robot that's close to a person, right?

00:04:20: The surface of that robot is not allowed to be more than forty two degrees because then you can touch it without feeling hot.

00:04:28: If it turns up something like fifty four degrees if your touched for three seconds You get a three degree burn.

00:04:35: That's big issue.

00:04:36: and this why efficiency in those systems are super critical.

00:04:42: But how we could go further with these system.

00:04:45: So the second issue is the size, right?

00:04:48: so you have two things where You have a very small size and you need to get the thermal away.

00:04:53: And this Is Where This Is Where It Starts To Get Super Super Tricky.

00:04:57: Also Where We See Customers Saying Well Standard Mods Fetch Yes!

00:05:02: You Can Kind Of Make it Work But Going For Ganfeds Actually Gives You A Little Bit Boost On Efficiency And It Startes To Be Worth.

00:05:10: While This Price Versus Feature Option trend moving more and more to GANFETs, simply because you can achieve that a little bit higher efficiency.

00:05:19: Because of the function of the GAN FETS And also because of that higher efficiency.

00:05:24: Also The losses on this system gets smaller so You Can build A Smaller Design.

00:05:28: So It's Kind Of All Coming Together.

00:05:30: The Same Is Also With The Processors Right?

00:05:33: the processors are starting to get smaller and smaller in package.

00:05:36: And this, of course is enabled because the nanometer process it's getting smaller and smaller so kind.

00:05:42: all these things coming together where you can say I can minimize my system that everything starts fit together but comes back from before when i mentioned processor communication to main controller as Thomas said.

00:06:00: Second thing here within coder interface.

00:06:02: It's not just one interface it can be two or three because you need if You have a gear, you haven't to have the encoder angle on both sides.

00:06:09: If you have functional safety potentially want a third one.

00:06:13: There is many small things you need to think of when you do that system for the current sensing?

00:06:19: It's a similar discussion.

00:06:20: I think we get back to a little bit more in detail.

00:06:22: but The current sensing you also need to have to write peripherals and the processor to get the right data from their current amplifiers.

00:06:30: All of this comes together and is where we also try from TI to show how can you make the smallest system?

00:06:36: And still give your performance.

00:06:39: Yeah, especially with the new type of robots that are co-bots where you have now a decentralized architecture.

00:06:45: very often I mean in the past for the heavy load robots You had a centralized architecture to get rid off.

00:06:51: The heat because of high power consumption.

00:06:54: It's difficult too at the heater way inside the joint after robot.

00:06:59: but this smaller type of Robots Customers are moving more and more to a decentralized architecture having the complete motor control inside the joint.

00:07:07: And here it's really important that you get rid of heat, then also have high efficiency in the gun technology for example with the high-efficiency is good solution from TI for coboards and decentralised architecture.

00:07:22: Also the microcontrollers with smaller sizes.

00:07:25: now they're getting smaller and smaller than you can directly put them Inside.

00:07:30: It's also something what is getting more important for the cobot type of robots.

00:07:35: and Also here then you have the industrial communication because it's a decentralized architecture.

00:07:41: So unique kind of real-time communication to synchronize the motors And that really needs to happen in real time with low chitter.

00:07:49: Low latency, and the peripherals Here are processors and microcontrollers like with the ICSS subsystem The Industrial Communication Subsystem Indicitara process For example where you have the flexibility to implement a different protocols like EtherCAD, Provenet and so on.

00:08:06: So your very open here for your protocols.

00:08:08: also on encoder side You have communication of the encoders in the joint now And here the processors are supporting the type of encoder protocol like this and that Hyperface DSL safety The Safety Protocols if we want to have a safety encoder.

00:08:27: That's something here where our customers can be really flexible.

00:08:31: Before we talk about current sensing, I want to come back to motor.

00:08:35: what are the typical motor types seen in robots?

00:08:37: What do you see on

00:08:38: market?

00:08:39: so today?

00:08:40: now it is a BLDC motor and i like to define also as BLDC motor but sinusoidally wound motor.

00:08:48: So basically your two field oriented control with We have seen some use of brushed motors in the hands because of low power.

00:08:58: However, what we also hear is this is moving to the BLDC motor Because the brush motors for instance create so much dirt that you cannot use them In clean rooms For example.

00:09:10: So limit where your robot can be used And there's a clear movement and everyone moves.

00:09:16: But

00:09:17: do you see an improvement or new trends?

00:09:20: new ideas coming from the motor side?

00:09:23: I would say.

00:09:24: right now we see a wish to try and figure out how can i do the motor control without magnets, but this is still very experimental.

00:09:33: And let's say there isn't clear solution on the market and it also something that you will not really be used in production ready robots.

00:09:41: However There Is A Clear Wish.

00:09:45: I want to try and fix my supply chain, and magnets are difficult.

00:09:51: So if i can avoid magnet?

00:09:52: My life is easy.

00:09:54: Super!

00:09:55: But it's a discussion.

00:09:56: we see customers who're trying to solve now or at least very high-level research wise trying to figure out.

00:10:03: Is there an other options?

00:10:04: because... ...I wanted get rid of this dependency.

00:10:07: How does it influence you as T.I when we have discussions in the market like this?

00:10:12: So, This discussion actually from a processor perspective if your look at algorithm is very similar.

00:10:18: so for my chip perspective they haven't of horse power to do now.

00:10:22: that was problem and passed with lot.

00:10:24: these motor types needed more algorithms.

00:10:27: powers who are doing math around them starting be possible know its one.

00:10:32: those where ideas existed already Fifty years ago, but it hasn't been possible to him

00:10:40: or

00:10:40: fifty fifty.

00:10:41: Yeah

00:10:42: So all these theories they existed for a long time.

00:10:45: It's more matter of.

00:10:47: yes You knew roughly how to do it But it was not Possible to do in an economically feasible way.

00:10:54: There are some systems already using this type of motors today.

00:10:58: I think one.

00:11:00: Yeah, I can probably not mention that customer name but But there are actually solutions on the market today using this type of motors.

00:11:07: Not for robots in other systems.

00:11:11: And when it comes to current sensing because Christian already mentioned current sensing What kind of current sending is needed?

00:11:18: For the drive?

00:11:18: and what is TI's perspective?

00:11:20: their USP?

00:11:24: offers different options, how you can do the current sensing.

00:11:27: I mean we have ADCs and we have Delta Sigma modulators.

00:11:31: so in our MCUs specifically the real-time ones like C- two thousand or also the Sitara devices they already integrated ADCs to measure the current with ADC's And also to synchronize The current measurements because it needs to be synchronized through your PWM signals So when you did a current measurement.

00:11:53: On the other side, we are also offering Delta Sigma modulators so that you have also the option to do measurements with these kind of current sensing.

00:12:03: And here We also have peripherals integrated for the sync filter For example in the Sitara devices.

00:12:10: That's done in hardware So to offload the CPU.

00:12:14: These are the different options depending on your accuracy and where you do.

00:12:18: The current measurement, which type of current measurements?

00:12:20: You're doing shunt based or whole sensor-based band widths and enopsed accuracy what you need.

00:12:27: so customer has a choice here with type off measurements he wants to do

00:12:32: when it comes to sensing data etc.

00:12:36: AI can play crucial role in this whole application.

00:12:40: or am

00:12:40: I wrong?

00:12:41: I would say here, yes AI can play a crucial role.

00:12:45: There is a lot where you could start trying to estimate and get a better current measurement using AI so that we understand the current movement because at the end of today when you do sinusoidal control You also have a sinusoidally current in this system.

00:12:59: So if there's a rough idea what it should be And then you can model this very nicely Using those AI functions for instance can help you do torque measurements based on the current, because of the math around the motor control.

00:13:13: With the current you can actually calculate a lot of these additional information using AI which can help you better understand and diagnose your system.

00:13:22: so yes there is

00:13:25: Aim to go in this direction.

00:13:27: or is it more the aim of the robotics companies?

00:13:30: because I'd see different strategies there.

00:13:33: So some robotics company say we want to do that, This Is Our Part Because We Have A Value Edit Service Or Value Edit AI To Our Product?

00:13:42: Or Isn't Your Turf?

00:13:44: At The End Of Today It's a Split Right if the processor doesn't allow enough processing power to do it, or the right accelerators.

00:13:53: And this is where TI.

00:13:54: we are focusing a lot on the tiny AI accelerator which we put into a lot of these real time cores and they're already in market today with engines so customers doing, for instance arc detection for solar inverters.

00:14:08: Or you can do use the current measurements to measure if the bearing of the motor is starting to break?

00:14:14: We have already these engines and an example showing how to do this so that

00:14:19: we are offering The hardware for it.

00:14:21: then we offer the tools for its to train your models but we will not collect data from the motor and train them.

00:14:27: models provide this into

00:14:29: our microphone.

00:14:31: We

00:14:31: provide the infrastructure, STK, HCI Studio for example where they...

00:14:37: Yeah

00:14:37: exactly.

00:14:39: and then hardware accelerators like in a C-II device is neural network processing unit to offload that main CPU on higher performance processors.

00:14:49: we have also DSPs and matrix multiplication accelerators in hardware in it And these hardware accelerates offer the SDKs and how to use them.

00:14:59: But at the end, data collection trainings of the model on the model selection.

00:15:04: that all has to be done from the customer And especially then for the use cases For predictive maintenance.

00:15:12: I think That's coming more and more where AI is used.

00:15:15: We talk since fifteen years about protective maintenance.

00:15:17: we are not yet there.

00:15:20: No, I don't get

00:15:21: there.

00:15:21: Yeah now we have the tools We can collect a data that you can simulate and test it.

00:15:29: Yes, I think one thing here is also important to understand with the discussion right?

00:15:33: We as TI we built electronics.

00:15:35: Right, we don't build mechanical hardware And this is where we also do a lot of the collaboration.

00:15:40: we help customers build these electronic systems With our components and they bring the hardware learn together how to use the electronic in the best way for the hardware.

00:15:51: And this is a key thing also with their whole doing the diagnostics and all these things, right?

00:15:56: It is you need actual that mechanical let's say solution to actually get the right data.

00:16:06: Of course, there are companies that want to do more than others.

00:16:08: But at the end of day we try and give the guidance on how to use the TI part with all the tooling Thomas mentioned before right?

00:16:15: How can you use it make a smart system And this is what will be trying show also what would help customers.

00:16:21: so coming back to hardware because we talk little bit about AI machine learning You already mentioned the communication layers or the communication protocols we are using in the robot.

00:16:32: Can you please explain once again how does drive communicate with a robot controller at the end?

00:16:37: Thomas, you want to start?

00:16:38: Yeah.

00:16:39: It depends on the architecture.

00:16:40: I mean if you have a centralized architecture or decentralized architecture.

00:16:44: in the centralized architecture The distances are quite short and here You often also proprietary solutions.

00:16:53: when it goes then over longer distances To the arm of the robot like an encobots to the joints.

00:16:59: Then there is for communication And control.

00:17:02: you have the industrial ethernet communication protocols Like EtherCAD for example And on the other side from the encoders getting back to position information you have the encoder protocols like end art and Typically.

00:17:16: then those encoder cables go to the base of the robotic arm.

00:17:21: I'm from there.

00:17:22: They are aggregated together, and from there it goes.

00:17:25: Then we ease a cutback to the robot controller or to the cabinet that you don't have all The cabling from the six or nine joints whatever how many accesses?

00:17:36: route to the robot cabinet.

00:17:38: So saving cabling going then from the base of the robot via ether cut or other Ethernet based protocols,

00:17:45: you want something Chris?

00:17:46: Yeah and also it depends on how they're robot manufacturer built their software architecture right.

00:17:52: because at the end of the day if To do the robot movement You need position control but to do position control you need to do speed control And then you need Pending what the customer says that drive is supposed to do.

00:18:05: Is it only doing current?

00:18:06: It's at only doing speed.

00:18:07: and where does the precision lie pending Where you place these blocks in there either robot controller or in The Drive, it depends how fast Do we need to communicate And How much data do You Need To Communicate?

00:18:21: This is also with the recurring and using real-time protocols for this, right?

00:18:25: That's why you need the industrial real time protocol to be able communicate reoccurringly fast enough.

00:18:31: To choose where do want put your control what we typically see as speed control current in drive then position in main MCU which makes sense because position control, you need to align all motors with each other.

00:18:49: To get the right smooth movement Right?

00:18:51: The main controller needs to know All of data anyway.

00:18:55: so it makes sense that It also does the math for and then You can offload the MCUs.

00:18:59: but at the end Of today is a customer choice.

00:19:02: there Is no right or wrong right.

00:19:03: its A choice.

00:19:04: And There are reasons For doing in one Or the Other way.

00:19:07: And not only the robot controller, but also a safety module needs to know the position of the joints.

00:19:13: and Here it depends then on the architecture how you route The cabling or your communication protocol because the main controller Needs it in real time.

00:19:24: But all sort of safety module.

00:19:26: It's kind off in real-time To be able to react fast enough in case if something goes wrong.

00:19:31: I want to come back because our episode is about motor, current sensing and power stage.

00:19:36: What kind of trends do you see when it comes to powerstage?

00:19:40: So if you look at the power stage what we see as a trend is definitely for cobalt humanoid.

00:19:46: It's much talking about miniaturization and high efficiency And using the technologies to achieve the highest efficiency.

00:19:54: Of course there

00:19:55: was always the goal right.

00:19:57: Yes yes but With the humanoid, it's gotten more extreme on how small he needs to be.

00:20:03: There is a second goal that starting to also appear in these discussions.

00:20:07: It's also if I want to mass produce it How much?

00:20:10: Is it allowed to cost?

00:20:11: right?

00:20:11: and this is exactly where for instance technology like GAN FETS they are In a price range Where there are starting To Be Interesting but still maybe A little bit expensive.

00:20:22: But what we Also see with GANFETS The More They Get Produced The Cheaper They Will get And Eventually It's not a question anymore.

00:20:29: The additional benefit they bring you is worth it because there are very similar pricing with the MOSFETs, so here we definitely see a trend of customers testing more and more GANFETS Because They See The Benefits They Bring.

00:20:43: But They Also See The Price Is Going So Much Down That Also Commercially It Makes Sense And Then Its Not A Question Anymore.

00:20:51: The GANfets Basically Bring One Caveat To It Very Fast Slue Rates Switching Times And this you need to also compensate.

00:20:59: when then, that your current sensing is not affected by these high slew rates.

00:21:05: This where we have done some analysis from TI with our different currents and technologies.

00:21:09: so trying Hall-based, trying shunt based measurements with Delta Sigma modulators with High VIN input or High common mode amplifiers... ...and trying to compare and see what does fit in the best way?

00:21:23: That being said, right?

00:21:24: We discussed a lot here.

00:21:25: This is all low voltage I would say sixty volt probably up to one hundred volt type systems.

00:21:31: we also provide gate drivers with basic and reinforced isolations for the standard robots.

00:21:37: Here The discussion Is more a price of.

00:21:39: i need To cost optimize And make this system More affordable simply because that market As mature.

00:21:46: you roughly know what You Need to do but now you need to architect A System discussion on issue in the past.

00:21:54: So a lot of times some of the solutions we see today, they are over designed because you got the price you've got right?

00:22:02: And this is not the case anymore.

00:22:04: there is a prize competition now and the functions are very similar on their robots.

00:22:07: Right

00:22:08: Yeah another trend I think it's the regenerative braking.

00:22:12: so In the past You had your breaking resistance.

00:22:15: Does

00:22:15: that make really sense?

00:22:16: The discussion doesn't really makes sense.

00:22:19: Yeah, I mean you want to save the energy what?

00:22:22: There

00:22:22: were tests.

00:22:23: I think it was ten years ago.

00:22:25: they tried do that but i said oh It doesn't pay out at the end and is now in this stage That They Want To Do That Really In Big Scale.

00:22:35: So If You Look At The Mobile Robots They Already Do Today Because The Mobile Robot Are Carrying so Much Load That The Energy The kinetic energy that robot has in the motors is enough to make sense, right?

00:22:48: So here it's already done.

00:22:50: We also see for industrial robots bigger ones where they move a lot of payload and do so today.

00:22:57: Where we can see at this moment with lower-and-lower pay load... ...the amount you have stored in the system is less.

00:23:06: But these are also what we as TI try to show.

00:23:10: You don't need to build that.

00:23:11: It doesn't have to be expensive, we can do it cheap.

00:23:13: We already have an option where we could use one of our power chips To regenerative energy and store energy locally.

00:23:21: Look at the DC link and put it down without any processor.

00:23:25: This is all possible today because ten years ago That little energy is not worth the cost that it

00:23:33: takes.

00:23:34: This is starting to change now, because we have also been looking into and trying to understand what's a real problem?

00:23:39: And how can you solve it as cheaply possible in the smallest way which isn't allowed at all!

00:23:46: So this is an advertorial episode.

00:23:48: right but please summarize in three sentences.

00:23:52: What does the whole offer from TI look

00:23:56: like?

00:23:56: I mean, we are offering products like processors sensing parts ADC's and we offer the software for the processors.

00:24:05: We offered a tools to program the processors and then in addition what our systems teams are doing, we're also doing reference designs.

00:24:14: To showcase different motor control applications.

00:24:17: for single-axis motors We have reference design for dual for upto six axis actually Reference Designs And were having specific reference designs where we show their capability of the different technologies like with gun technology For example there is a reference design available.

00:24:36: So that really helps the customer to get started and test the performance of the products from TI, how they play together.

00:24:44: Christian wants you add something in the product range?

00:24:47: Not ad but also show.

00:24:48: right.

00:24:49: we were at CES this year And here We showed a robotic arm where we showed The joint with reference design Tomas mentioned Right With GAN Small form factor showing EtherCAD communication very small form factor for the hand to show how do control with one processor six motors.

00:25:11: And on that demo, we also showed new technology from TI working on a torque sensor and this is still in sample prototype so it's early but we showed at CES And then a second

00:25:24: was the feedback.

00:25:25: The feedback is super interesting technology, very interesting.

00:25:29: customers want to learn more see how they can use it for different use cases and this Is what we are starting now?

00:25:34: This effort A Second Technology will also so as a tactile sensing where We show How To Use The Whole Sensors With Magnets On A Material In How Sensitive This Is.

00:25:45: If You Look At That System in The Demo Then you see that this has a very nice sensitivity.

00:25:50: Of course, there is the problem around... You put metal around the magnet and it will distort the fields.

00:25:55: but here again That's where AI comes in right?

00:25:57: The AI can model this out potentially And then you have an working system.

00:26:02: Yeah, TI is kind of a one stop shop.

00:26:05: I think for customers because they get everything that you need

00:26:08: to order whole robot here now know

00:26:10: not the whole robot but from their electronic side i mean From the semiconductor side.

00:26:14: You Get all The Analog components?

00:26:16: You get your digital components with MCO's processors scaling portfolio Here and you get also on the software site support here With the SDKs Your real-time operating systems the AI tools plus the industrial communication protocols.

00:26:33: So that's also available from TI like EtherCut and so on the encoder protocols, and that is all supported in the SDKs.

00:26:41: actually

00:26:42: I already asked Christian do we underestimate it?

00:26:45: And he didn't answer really The question.

00:26:48: Thomas Do we under estimate hardware at the moment From your point of view

00:26:53: for the industrial robot or further for the drives

00:26:56: for the drive for the whole robotic system?

00:26:58: i would say

00:26:59: On the electronics side, I think there are still a lot of challenges like thermal problem efficiency and also the size.

00:27:09: Those three topics are underestimated especially for new co-bots type of robots.

00:27:16: customers have problems to get rid off heat here.

00:27:20: For more mature industrial heavy load robot I think they are more working on the cost optimization and that day run twenty four hours, seven days.

00:27:31: So with a predictive maintenance now already detecting if there is a problem before the robot fails.

00:27:37: actually.

00:27:38: so it's something.

00:27:39: what?

00:27:39: What do you're working on?

00:27:42: know your want to answer?

00:27:43: yes And okay um...so..what i would say Is-I Think Yes It Is Underestimated.

00:27:50: And I Think If You Look At any company in the world today.

00:27:53: You have maybe ten hardware engineers for the full system, but you'll have thousand software engineers exactly and this at the end of day also always brings some let's say, issues on what do you want to achieve?

00:28:08: How much electronics through your wall to build in house.

00:28:10: Do I need to buy it externally?

00:28:12: and this is also why we see a lot in the companies that they have to by externally because they don't have the people who actually due their electronic design.

00:28:20: And there's also where we try and help them too make the speed of an electronic designed faster By helping with layout reviews Helping with schematic reviews helping With electronic understanding Of discussing How does the system work?

00:28:35: Also, how do we think it works?

00:28:37: but also.

00:28:38: It really worked because the customer always knows how it really works.

00:28:41: We only have assumptions or assumptions are typically pretty okay But sometimes you miss details and this is where I would work with the customers right.

00:28:49: And here definitely there is a tendency to let's say underestimate The time it takes.

00:28:58: industrial ready system because

00:29:00: great.

00:29:01: Exactly, Because there is a difference.

00:29:03: saying I built an electronic system and let it run in the lab?

00:29:07: Yeah this is pretty straightforward But as soon you now put it in an industrial environment,

00:29:12: with a lot of noise.

00:29:14: Exactly!

00:29:14: With all the noise then suddenly communication will stop working because design was not done properly.

00:29:19: You didn't do the right termination or layout In this system and ground planes to get the return path.

00:29:29: All these small details.

00:29:30: that does different work?

00:29:33: And this is definitely something we see.

00:29:34: I think one good example for this, single-twisted pair ethernet.

00:29:38: what?

00:29:38: We see single twisted pair Ethernet for automotive.

00:29:41: the way it's done works nicely.

00:29:43: as soon As you take that system and put it close to a high voltage noise source with Determination they chose

00:29:50: experiment on that right

00:29:52: not experiments.

00:29:53: it's already working solutions.

00:29:55: We have all ready test reports showing how do you pass EMC for industrial systems?

00:29:59: Okay, not a thought experiment anymore.

00:30:01: It works and they're already customers doing industrial graded systems with single twisted peer.

00:30:06: I just don't the automotive way because this is not noisy immune enough to abyss stand an industrial environment.

00:30:13: And these things are important to grasp on understand them.

00:30:16: I think that there whole humanoid and copot was added.

00:30:20: less weight.

00:30:21: single twisted pair ethernet will definitely be more and more important.

00:30:25: And it will be a requirement to make low weight robot system, and try minimize the cables.

00:30:32: A second thing is diagnostic discussions that Thomas mentioned.

00:30:36: The diagnostic discussion with Single Twisted Pair Ethernet.

00:30:40: There's something called an SQI value inside.

00:30:42: You can read out of the registers of the PHY Which basically sees what signal strength comes from other PHY during runtime.

00:30:50: So you can actually see what is the signal quality at any given time.

00:30:53: If this starts to go down, connector cable something's going bad I need to do something before it completely dies and all these things also small micro steps that are starting turn the needle more.

00:31:07: giving options were not available ten years ago.

00:31:09: Christian already shared his outlook.

00:31:12: so Thomas your outlook when it comes to current sensing the whole robotics field.

00:31:19: What do you see, what we will see in the next one to three five years?

00:31:26: I think that predictive maintenance topic is one thing which Really now, yeah.

00:31:32: Especially in combination with AI and all the possibilities here what you have to get more information out of the motor or system that can react up front before it really fails because your band is standing still... Factor if the system breaks something can go wrong in your factory.

00:31:55: So I think with predictive maintenance coming more and that

00:31:59: problem Can we reduce sensors on the robot?

00:32:02: I think we might have even more sensors in the future because of the safety topic.

00:32:08: We

00:32:08: will have a next episode on safety, right?

00:32:10: Exactly where we'll cover that topic.

00:32:13: TI is offering here radar sensors.

00:32:16: also on the safety side there should be maybe even more sensor.

00:32:21: But

00:32:22: the idea from AI was to reduce sensors or am i wrong?

00:32:28: Yeah, but the use case is getting more and more in the direction that industrial robots or co-bots are working next to humanoids.

00:32:35: So I mean the safety standards... Next to humanoid?

00:32:38: What

00:32:39: do you say?

00:32:39: Human!

00:32:40: But

00:32:40: this was a next step Thomas

00:32:41: right?

00:32:41: Co-Bot's

00:32:42: Working

00:32:44: Next To Humanoid.

00:32:45: In the past we had the cage around where it used to be a place where the robot only worked but in future with collaboration between humans and robots The safety topic comes and there you need more sensors.

00:33:01: in the past.

00:33:02: You just had a fence around it.

00:33:06: Yeah, on top of this one that Thomas mentioned right with the co-bots they are already able to work next to humans but when they're working so slow then they aren't economical.

00:33:19: And this is where the mention from Tomas about how better you can detect.

00:33:26: How far can we go?

00:33:27: Yeah,

00:33:28: let's say how close is the person allowed to go before?

00:33:31: I need to slow it down.

00:33:32: And i think this is The next step where the efficiency of those robots will increase because Of all the sensing and with AI you Can actually take the data and process It quick enough that You can use it.

00:33:44: That was the problem Before and This Is Where this...I don't Think We Will see less sensors..i think we'll See more but the ai will be able To Take the Data and Actually Make Use of it quick enough because you need to do it real time, right?

00:33:57: But we will talk about safety and centers in the next episode.

00:34:00: Thanks a lot Christian.

00:34:01: thanks a lot Thomas was a pleasure again.

00:34:03: Thank You

00:34:04: Robert.