In Between the Tech: A Conversation with Sneha Sudir Shetiya of Torc Robotics: The Tech Between Us

In Between the Tech - Zonal Architecture 
Raymond Yin 

Welcome. You're listening to our subscriber exclusive episode of In Between the Tech. We are here today with Sneha Shetiya, staff software engineer at Torc Robotics who joins us to share her thoughts on the innovation surrounding Zonal Architecture. Sneha, we're excited to hear your perspective and learn how this technology will drive change in automotive design and development.

Host 
Thank you for joining us today on In Between the Tech. You've been in the automotive space for a number of years. Can you tell us about your background and Torc Robotics?

Sneha Sudhir Shetiya 
I have been, for a couple of years now, in the automotive space around seven plus years. I started my career in powertrain with engine and aftertreatment systems. This was way back in India in 2014. I worked for a couple of years there and then came to US for my Masters and completed my Masters in computer vision and signal processing, and that's how I transitioned to ADAS vehicles. That is autonomous driver assistance systems. And for a few years now I have been working in that domain. So that's my brief experience here. As far as Torc Robotics is concerned, we are an independent subsidiary of Daimler Trucks. We provide level four autonomy for trucks. That's our major business use case to provide hub-to-hub transport. We definitely try for hands-free driver, but along with all the necessary safety measures, of course, that's our main business use case. So here at Torc, I'm basically part of their embedded trucks team. I'm focusing on the diagnostics of the sensors and the platform software.

Host 
In autonomous vehicles. How are neural networks utilized in their development and how do you see this affecting the automotive industry as a whole?

Sneha Sudhir Shetiya 
Neural networks, definitely they play a very pivotal role, not just in AI and also in the automotive industry in the ADAS domain especially because their work revolves mostly around all of the sensors that is cameras and radars and LIDAR. Basically, LIDAR is nothing but the laser detection and ranging, that's what it stands for. They perform two critical tasks for us. First is the object recognition, identifying the object on the roads such as pedestrians, vehicles, lights, and all of that. And also understanding what those objects are doing. That is called a scene interpretation. So what is it about understanding the context? How are they in harmony? So neural networks basically helps us to understand then the relationships between these objects. And they also help us to predict the patterns of the vehicles. How are these objects moving and what are the dangers? That's how it'll enable the vehicle to become autonomous. So basically coming to the automotive functionality of it in the car considering, or a truck, because my industry is truck, I'm going to put in that there. So it's basically utilized for the steering and the navigation. So it will follow the traffic flow, the road signs, and also the environment understanding. Basically to understand what's the optimal course of action it has to take. And then this will enable us for a smoother journey, a safe journey. And also, it also helps us to understand when the autonomously they have to apply the brake to avoid a collision. Also, it helps in road planning. When traffic is there, how do you avoid condition? So neural networks actually factors in the real time traffic data. It factors in weather conditions also. Today we have data sets which determine weather conditions, night and day, time data sets, and also it helps us to understand when to stop the vehicle if you have to do a drop off. That's the major use case of all the robot taxis we see around. So that's the major impact I would say. I think eventually as in when the software gets more mature and better, it'll actually be very safe as well and found to increase efficiency over general human drivers. So because it becomes like traffic and everything, it becomes additional chore for some drivers, some drivers which are disabled or all of that physically unable to drive and stuff. So all this use cases actually come into real use case. I guess

Host 
Those are interesting scenarios and we are excited to see those developments in autonomous transportation. Let's talk about advantages in zonal architecture design specifically in relation to sensor fusion. How will this help create the next generation of driverless vehicles?

Sneha Sudhir Shetiya 
So before we go to zonal architecture, we need to understand what are the cons of centralized architecture that we currently have in all of our standard internal combustion engine vehicles. What is happening there in centralized, all of the data is going to a single ECU or a single computer, you can say like that. So that results in a few problems. One of that is the data overload. We get overwhelmed, the computer or the ECU that you can say gets overwhelmed with all of the sensor data that's coming in, which leads to more processing time. Then the wiring itself becomes very complex because then all the sensors are connected to one single unit. How much of maintenance you have to do and the weight that it increases comes with it. So complexity of the system is increasing. And also this kind of system, when you design, it's very difficult to scale. So when we talk about freight or cars, it's not just one or five, but in terms of thousands, thousands of cars that you are doing it for manufacturing. So all this are the issues that are found in centralized architecture. So from there is why the need for zonal comes into picture. So zonal basically it helps in sensor fusion in a way. Is that the zonal controller that's there, right? It is basically divides it into zones like a front zone, rear design, and you can divide it into various zone the car chassis, and then it gathers the data from various sensors based on the zone it is, and it only provides processing for that particular zone. This helps to curb the first con, which I said to avoid the data overload on one ECU, and then also the decision making is localized to that particular zone. In that case, because it's already divided and wiring, the complexity of the wiring also reduces by a significant amount. So I think these three main things add on for zonal architecture in autonomous vehicles.

Host 
You mentioned ECUs. As more and more ECUs are added, the complexity grows in vehicle architecture. What are the considerations designers are addressing in systems like ADAS development?

Sneha Sudhir Shetiya 
Yes, so majorly, I think when there are many issues involved, a major concern comes in when you're doing the final integration, right? Okay. Each module is developing its own software. Every team does it, but finally everything has to come together and work as a system for the ADAS to work. So this involves a lot of data exchange between the issues that has to be very seamless, and we have to also have data synchronization compatibility of the communication between these ECUs so that malfunctions don't occur later on when actually they're on the vehicle. So these things we have to take care of. Although it looks like a hardware topic, it does have a lot of backup in the software management side as well because each ECU has its own software and then you have to do maintenance of it. You have to update it, you have to debug the software and this has, when you do all those things for one ECU, it has to work in harmony with the rest of the ECUs. So this can be very much time consuming and error prone for the CI/CD people who work on the software. So software management is very crucial. And then also power consumption. So as and when go on adding ECU, each ECU will consume, add on to the overall power that is consumed. So this adds lot to the efficiency aspect of it. If it's fuel based and fuel efficiency and also electrical system will get a lot of load apart from data. And also sometimes it can cause safety concerns because then multiple ECUs are actually handling the data's functions. And then we have to ensure that every ECU is safe and all of them as a system are also safe. So in order to curb this, how we can do efficient design is, one is we have to standardize how you are configuring each ECU, how the communication is happening. So, if you have standardization, then they will work very much in harmony because in older ones we do have like these kind of standards and all of them, which we can leverage here to utilize it. And then the virtualization power. So virtualization in the sense maintaining a common interface that multiple ECUs interact with, so that additional communication overload is not there. Design also should be very modular, very smaller, so that debugging is easier. And prioritize. So which ECU tasks are prioritizing? Which features do you want to prioritize and which are doing the same things? So there may be different aspects of sensor fusion where object detection is being done by one ECU, the same may be used for localization in some other format. So these are all redundant tasks. So you can always reuse some of it so that you're not, again, increasing lot of software on the system. And last but not the least, this is always important, let it be ADAS or connected is focus on the security aspect of it. So nowadays we listen so much about cyber attacks and all of that happening. So we need to have enough investment to have enough encryption algorithms built into the system so that security is not at all compromised in this kind of system.

Host 
What role does edge computing play in transitioning over to zonal architectures in vehicles?

Sneha Sudhir Shetiya 
Edge computing actually it gets the data closer to the sensors, basically. So it is not very far away. So what it happens by doing so is real-time processing. So edge computing it enables - so we have already now divided the car into zones with each having their zonal controllers. So, what now it is enabling is to process all of the zonal data in real time. So this enables us to make faster decision for the ADAS algorithms, which this data is being used for. So warnings are easier break warnings, emergency brake warnings are easier. So that is one advantage. Second is because the data is very closer, now, the filtering is quicker and pre-processing is very easier because when we talk about neural networks and ADAS features, data plays a very, very huge role. And these edge devices, they actually help in filtering the data and pre-processing the data. So this needs to be sent to a central computer for further analysis. And after that, the bandwidth usage and also helps in reducing the processing load on the central unit. And also it helps in a way determine the security also, right? Because then all the sensitive data processing, which is close to sensors, it can be done through the edge devices and thus it can have data transmission across the vehicle network and thus also reduces potential severity risks that we see with security. So I think, yeah, this would be a good understanding.

Host 
You mentioned this a little earlier, but how does zonal architecture impact vehicle diagnostics and maintenance procedures when compared to traditional domain architectures?

Sneha Sudhir Shetiya 
Yes, so this is very good topic because it's very close to what I work on also on a daily basis. Because diagnostics is a pain and especially it's a bigger pain for ADAS features because earlier in powertrain we just have two ECUs. One is doing the engine functionality, one is doing the after treatment. We do have the chassis thing, but in ADAS itself, all the sensors, every sensor acts as one ECU. And again, the main platform ECU, which actually houses the software. So zonal controllers, they provide a very consolidated view of sensor data and system health within their zone. So this enables us for faster and more precise fault identification. So faults is something which is very an entity of diagnostics, I would say, and that is what we use to diagnose the problem. Diagnostics is nothing but the troubleshooting. So identification of fault comes primary and also diagnostics is done over interfaces. So if we have different kind of interfaces, then we need to have new diagnostic techniques for different interfaces. So zonal architecture helps us to standardize these interfaces because it uses standardized communication protocols across these zones. And this helps in minimizing, simplifying the software updates and also the maintenance procedures. And then technicians, we can use a single tool for these standardized interfaces to access and update the software process for the ECUs in that particular zone. Apart from that, it also helps the zonal in things like secluding the faults in each of the area. So fault isolation kind of a thing where it helps to isolate the problems for that zone. And then you can spend less troubleshooting time because you have only faults of that zone to find out instead of the whole vehicle having the complete problem. So, what are the benefits of this? I would say basically is the downtime is reduced significantly, efficiency has gone up. There is system transparency because all the zones are there and we know the faults that are in those zones so we can diagnose them better. Tools are very standardized because of standardized interfaces. And one more important feature, which nowadays ADAS utilizes and every connected vehicle, is the over the air updates, the remote software update, which many of our cars have it today. So zonal architectures actually help in that. So you don't have to take it to the technician every time there is some software failure, you can update it at your home using the zonal architecture feature that is over the air updates basically.

Host 
Can you tell us more about the implications of zonal architecture on vehicle power distribution and management?

Sneha Sudhir Shetiya 
Yes. So conventional car architectures, they mostly rely on centralized power distribution, and the battery basically is supplying power to one junction box and then that is distributing it to various issues throughout the car. So this is our more traditional approach. What are the problems again of this approach? It is again, the same wiring complexity. Lot of cables are needed, thick cables to the central box, which is providing to the ECUs. And because of those cables, the weight again, is increasing of the vehicle, apart from all the other things that we have going on there. This again, architecture is not scalable because adding new features or ECUs again, enables more cabling, more heavier wires and more weight. So thus it increases the weight and increases the capacity and just is not scalable, I would say. So there is, again, single point of failure. So there is one malfunction and in the central junction box done, everything is gone. You have to replace the whole power distribution system and if it is some critical system that you lost power in, and then that can be really fatal. It can cause accidents and stuff like that. So definitely this is not the way we want to go. So we go zonal architecture. Then again, we power to the zones only so it decentralizes the system. So what it does is we have already divided into zone, like the front end, rear end, cockpit in all of them, and it provides power distribution modules to each of these zones. And that power distribution module provides power to the ECUs of the zone they are presented. So this again minimizes all the risks that we talk before with respect to wiring. The scalability aspect is better because you can put more PDMs, the more number of ECUs you add, you can put in more PDMs as well, and reliable, more reliable as well.

Because again, you can isolate what is the fault and we can ensure we give more power and very good backup thing for the critical systems. We can isolate all the critical systems in one zone and give it a power to that zone. So these things, we can do it utilizing zonal architecture. We can also go a little bit step further by enhancing using smart fuses like we have it in our homes today where we have kids and stuff going around. So we all have smart fuses, smart home, so that kind using different material. I think they use MOSFET for it, basically a metal oxide because the field effect transistor, it offers more control over the power delivery and it can respond faster also in case there is an overload, power overload, and it also protects shorting of the components. You can regulate voltage better in these kind of power distribution models within each zone. And you can do a battery management as well better because it allows for better integration of the battery management systems. So nowadays EVs have these features like regenerative braking and stuff like that, right? All this is enabled because of better battery health monitoring and all of these features. So yeah, definitely it's very much advantageous, I would say.

Host 
You brought up an interesting point about the complexity and weight of wiring. How does zonal architecture contribute to reducing the complexity and weight of wiring harnesses in modern vehicles? And are there any other advantages worth noting?

Sneha Sudhir Shetiya 
Yes. So I would say zonal architecture follows the principle of divide and conquer kind of, I guess. So the first half divided the vehicle into each zone with each dedicated processing unit for zone and each PDM, the power distribution model that we just talked about. And then this enables shorter run times, right? Because with zones and PDMs, you don't have to run longer, the cables don't have to have been very longer. Thus, it also increases the weight, decreases the weight, it gives a very modular design because it has developed into zones and there is a very redundant kind of an architecture you can develop. So one zone you develop with one PDM, maybe three ECUs, then you can restructure the other zone on similar ideas. So it is a very modular design approach. So I think these three are the main things that it enabled.

Raymond Yin 
We hope you've enjoyed this exclusive in-between the tech episode for our subscribers, thank to Sneha for sharing her expertise on zonal architecture. And thank you to our subscribers for supporting Mouser. You can explore more of our zonal architecture offering, including articles, videos, and more by visiting mouser.com/empowering-innovation.