Here is the transcript for the webinar “Connected and Automated Vehicles (CAVs),” presented in May 2023 by the Vehicle Technologies Office of the U.S. Department of Energy.
--
Kevin Viita:
Presentation cover slide:
Hello, everyone, and welcome to the fourth in the SMART Mobility webinar series brought to you by the national laboratories and the U.S. Department of Energy. My name is Kevin Viita with ITS America, and I'll be helping facilitate today's webinar. Before we get started, I'd like to cover a few of today logistics for today's webinar. We are recording this webinar, and the recording along with the presentation slides will be available at a later date from the Office of Energy Efficiency and Renewable Energy’s website. All of the attendees are muted by default, but we'd like you to stay engaged by using the Q&A pod for comments and questions. We plan to have a Q&A session at the end of the webinar, so if you have any questions during today's presentations we encourage you to enter them as they come to your mind at any time. We have two presentations today from across the national laboratories.
First presentation slide:
First we'll have Dominik Karbowski, group manager of intelligent vehicle control at the Argonne National Laboratory, and Yunli Shao, R&D staff member at the Oak Ridge National Laboratory.
Next slide:
That's all I have, and with that I'll hand it over to our first presenter, Dominik Karbowski to start us off.
Dominik Karbowski:
All right, thank you, Kevin, for the introduction. So today we'll discuss about connected and automated vehicles, looking first from the vehicle and powertrain control side and also another traffic lab.
Next slide:
So we'll show how driving automation and connectivity can be used for not just safety but also energy efficiency as well. But first maybe let's start with vehicles that have automation and connectivity and which are already on the road today.
Next slide:
So here's one example of a connected automated vehicle. This is part of a data collection, over 2,000 miles, two different – a level 2 production automated vehicles that were tested. The data is by the way available for everyone to look at. So this little video shows what happens when there are cut-ins and cut-outs. And what we noticed from the data is that there's actually significant delay in the response of the automated system, and that causes some disruptions, dips in vehicle speed, that may not occur with a human driver. And these delayed reactions may cause penalties for energy efficiencies. So probably faster perception stacks or better object pass forecasting would help alleviate these.
Next slide:
We've also looked at the sensors themselves because they're an essential part of the system. And looking at how the nominal specification compared to what you see in field testing. So we'll look for example at LiDARs and cameras. We know in the LiDAR we tested, we noticed that the object detection is 75 percent lower than the specified nominal range. On the camera side, we outfitted the camera with different algorithms, object detection algorithms, and there's actually quite a gap in their performance. So not only the algorithm are very important in the performance of the systems. So those are factors that modelers, OEMs, need to take into account when assessing performance and control robustness of CAVs.
Next slide:
Another technology that's out there is sometimes called GLOSA for Green Light Optimized Speed Advisory. So this works as a speed advisory to the driver when they approach a traffic light. So you can see here, or in that circle you see that speed advisory that's displayed. If the driver pulls that speed, he will go – he or she will go through the traffic light without stopping much, and in the most efficient way. Why we actually see and practice is that this system does not see that there are queues before the traffic light. So when it sees, for example, at that point that 45 miles per hour but there's a vehicle that is just in the queue and preventing it from moving. So either the infrastructure who should estimate it could estimate or detect some of the queue links for the speed advisory, and maybe the closer systems could also take into account the sensor information sensing preceding vehicle so you could adjust these advisors.
Next slide:
We don't have your truck platoons commonly on the roads but there's some experiments out there, and our colleagues at NREL look a bit how – learn a bit from real-world experiments. Here we can see that there's a cut-in between two platooning trucks. It's fairly short. But that short event that lasts a few seconds actually creates the of the platoon to this band and then the reformation of the platoon takes a long time, increase fuel rates penalty. So if we consider for these type of future-looking technologies we should definitely – or stakeholders could really look into how we can discourage cut-ins in and maybe to prevent these things from happening.
Next slide:
We look a little bit at experiments, at experimental data from control experiments. How about fleet level? How about when people use it in the real world? So we partner with a fleet and look at data for over 1 million kilometers’ worth of driving data, 40,000 trips, and about 160 vehicles total. We looked at how often ACC is used. So adaptive cruise control, which is a backbone of many – of any other automated driving vehicle. So we analyzed. We found out that on about a third of the trips have ACC on at some point in the trip. From that we can develop ACC engagement models. Actually we can find out that in this particular fleet they're very, very well-related to the road link attributes. So you have, you know, if you know for example, if it's on a highway there's a much greater propensity to turn the ACC from the model. We also look at energy impacts. So we looked – we use regression testing but controlling for vehicle type, temperatures, great speed limit, and/or a range of other factors that affect energy efficiency. So our conclusions were that ACC in that fleet increased fuel consumption by approximately two percent. So one reason – and that kind of confirms what we've seen in previous slides, where the systems have still room to grow in terms of energy efficiency. And part of the reason is that maybe there could be more. This objective, which may be not the original objective of automation, could also be included in this. And we'll see later, and we'll discuss that, but standardized methods and scenarios to really evaluate these would be helpful. So if you want to, you know, if you want to address these shortcomings of current connected – automated connected and automated vehicles, how do we go about it?
Next slide:
First we need a workflow of tools, a toolkit that helps us do development and research efficiency. Deploying, building and deploying, and testing these vehicles is challenging and expensive. So we want to start in simulation. We can run a lot of different scenarios. Of course, one key is to have very – as high fidelity as possible of a simulation as we can. We can, of course, we can look at – we want to target, to do track testing and then potentially on-road with real systems. But in between simulation and these real system testing, we use a technique. At those labs you will see today use this technique XIL, which stands for anything in the loop. So we have that acronym. It's really a mix of real and virtual systems interacting with each other. For example, you have a real vehicle that is on a track, but it's linked to a digital twin, a virtual world that tells it that, well, there's a traffic light coming, there's a vehicle – there are vehicles ahead of you. And in that way, we can test certain scenarios in the safety of a controlled environment, and before we move to more challenging environments.
Next slide:
Science, we need representative scenarios. So you know, if we wanted – if we – if for regular cars you have your MPG sticker, right? That MPG sticker is calculated when we – through going, putting the vehicle for a series of what we call drive cycles, duty cycles. So they're predefined. There's two or three of them. And they run for that and we have the number. It's much more difficult for a connected automated vehicle system, because the way the driving itself may change based on the technology, and that's actually what one key feature. And we see that it changes a lot – the energy impacts can change drastically from one scenario to another. So we need a way to create these scenarios. So what for example, one approach we've taken is looking at, let's say if you want to model with representative scenarios for entire metro area. We can look for example at data from Polaris, which is presented in a previous SMART webinar. With Polaris, they know pretty much where people go from where to where, what time and for what reason. So we can extract the most representative trips from this using machine learning and other techniques to have to down-select a smaller number of scenarios that we can run in simulation or in testing. So it's very important that that scenario standardization, it would be very important and would greatly facilitate the development of more our energy-efficient CAVs across all stakeholders from academia, industry, or labs.
Next slide:
So third, we need a human driver model. And you may ask, why do we need a human driver model? We're talking about automated driving. There's no human in there. Well, two things. First of all, we want to use it – we want to have a solid baseline for to compare any automated driving against. So that's one reason why we want to have a model. Secondly, for foreseeable future, connected automated vehicles will evolve in a environment with a lot of human driving. So one challenge is that a lot of the driver models were historically developed for traffic flow simulation. And so while they're very good at traffic flow, when you look at the, you know, sub-second-level type of dynamics they're not as realistic. So we've developed – we worked on developing a more higher-fidelity model using driving data from across the United States, one million driven kilometers. And that model is capable of better capturing the dynamics, the exploration, the high acceleration, the changes in acceleration, than existing models. So that's something that could be integrated when we look into com CAV controls development framework, whether the simulation or even testing.
Next slide:
So now that we have the tools, how we do – let's get into the meat of it. How do we save energy through driving automation? So let's just start from a simple example. Taking a short segment that is one kilometer, we go to a speed of about 30, 35 miles per hour. What's the best way to go from zero speed at one end to zero speed at the other end? So if we look, there's various – of course, there's multiple ways. In black line is – you can see the black line is a human driver, the model that we just mentioned in the previous slide. So that's our baseline. So if we use some optimal control theory, we'll find the blue line here is what gives you the lowest energy, traction energy. So that's what will be the most efficient. But if you look at it, strong acceleration, strong braking, cruising and coasting in the middle. So we also develop auto control that are a little bit more, let's say, user-friendly. So they tend to smooth the drive more. They also include cost function to minimize that are related to energy, but are a good – they're a good trade-off between increased energy savings and drivability. So automated driving just in itself, non-connected, can bring savings, and that's something that could be considered by stakeholders.
Next slide:
Now if we know a bit more, we can do more. The more you know, the better you can – the more you can see, in a way. And here if we know more about the powertrain, if we have controls that are a little bit more specific to the vehicle they're set on, they can save more. So we looked at having controls that can learn about how the powertrain operates, and then they can modulate the speed. So to keep the powertrain and it's at a good efficiency without penalizing travel time. There's also controls that directly can control the powertrain, the various pieces of the powertrain, such as the gear shifting or the power split in the hybrid vehicle. An OEM with powertrain knowledge and powertrain control, you can save up to five percent more than if you don't have that information. So that there's some trade-offs between complexity and energy performance, and when deploying these algorithm that's something to consider.
Next slide:
One also technique that can be used for improving the performance of automated driving is AI. So AI is widely used. I mean, it's the backbone for object detection in these automated driving systems. It's less commonly used for control. So we can use it to optimize the parameters that even after optimization are still left to the driver. So we've worked – we've applied this in an example, where we use reinforcement learning to find out what's the best way to go through a simple trip with intersection approach. And it can learn how to better prioritize the system and bring 10 percent energy savings. So definitely AI is something that could be considered to bring combined, maybe use some other techniques to bring additional energy and potentially travel time benefits. So we'll look at automation without connectivity. But connectivity for example, with to the infrastructure can bring energy savings.
Next slide:
So let's look at one case here. This is – let me rewind it. So in the next few slides we'll discuss about signal phase and timing information from traffic lights and how that enables CAVs to better anticipate, to better approach a traffic signal. So here's a demonstration of how this works. So you have a level 3 vehicle. So this is a work we're doing in partnership with University of Wisconsin, and these vehicles were developed as part of the previous DOE project. So they'll be approaching two traffic lights. And you'll see that the traffic lights have CV2x connectivity, and they broadcast their signal phase and timing to the vehicle and the vehicle has an automated eco approach or a kind of eco – sorry, a connected automated driving system, which works not just on traffic lights but more generally speaking. But here we'll focus on the eco approach. All right, so let me – I'll let you play it. Here we go. So you see the first green line is here. The vehicle is accelerating normally. And it will – the next traffic light is red, and as you can see, the vehicles start to slow down, because it knows it's red here, and it's kind of waiting for it to be probably green. And then now that it's green, it's accelerating. So this is an example of eco approach. This is a low-speed, more functionality-oriented test. We'll be doing more higher speed on track testing of these systems with multiple vehicles involved. So therefore SPaT broadcasting for eco approach a traffic signal is something that can save energy. So how much energy can it save? Let's see.
Next slide:
So here we run another range of experiments. So here on track XIL. So we had a real vehicle that was driven on the American Center of Mobility track, ACM, and that was done in partnership with our colleagues at Michigan Tech University. So several – so we could simulate the approach to the same traffic light at the same speed but at different time in the traffic cycle. So if you approach it when the light is green, of course, nothing happens. You don't stop. Whether you know the – you have this SPaT information or not, nothing happens. So there's no benefit. But when for example you approach it late in the green phase, so when you enter, it's green but it quickly turns to red, you can see the baseline case that does not have SPaT information brakes strongly and then stops at the traffic light, waiting for it to go green. But if you have SPaT, you can anticipate, slow down, and move on. And this KC is when you arrive at the red light. So not only you can save energy, 16, 25 percent just on these small segments, it also – you know, you save some time because you don't have to stop and go get back to speed. So that's something to consider for stakeholders, is that not only it can also have more reduced some of delays, as well.
Next slide:
Now if we expand that to longer scenarios, so here we show four different routes, so also on track and on dyna. So two different situations. And we also have different powertrains. So different powertrains may have – may react differently to this technology. So actually, the analysis shows that EVs save about the same as ICE-V vehicle, so conventional vehicles, or maybe sometimes more, from – so that's one finding that we can see on these routes. And that’s longer routes, so that accounts for some cruising phases. Although there's still, you know, they're fit – they're made to fit in one loop of a track, which means that maybe they include more traffic, more intersections, more simulated intersections that would be enough in more normal scenarios. So again, everything depends on the scenarios, hence the previous one about the need to have scenarios. But one point here to clearly make is that even in the context of electrification, adding connected, you know, corridors, or having information about signal space and timing, can still be beneficial and make EVs more energy efficient, so save on some of these Watt hour or extend the range of electric vehicles.
Next slide:
So we’ve looked so far – looking about information from the next incoming traffic light, but if you add information of more than one traffic light you can actually see even more. So here we have actually run a case study where we look at different range, communication range, and how it impacts the energy savings. So definitely having two connected traffic lights is a good trade-off where you achieve a higher energy savings and they – so these – of course, these depend on what's the maximum speed, the spacing of the traffic lights, you know, and you know, the powertrain, as well.
Next slide:
So we've looked more at individual vehicles or maybe a few vehicles, but if you know, eventually these systems, if they're – if we want them to be on the road, they'll – you know, we'll have some level of penetration. And they may have system level impact, so it's important to also perform the eval, the assessment, at the system level.
Next slide:
So we look here – so we insert it in simulation, in traffic simulation, these CAVs at various levels of penetration in a – you know, in simulated traffic. So the key takeaway here is that – so first of all, they can bring energy savings, even a low penetration rate. So which means that non-connected vehicles benefit from the connectivity of others. So for example, 20 percent energy savings at 30 percent penetration rate. And also this type of technique does not, you know – if it's properly calibrated and does not affect the throughput of the corridor, does not have any negative impact on traffic. So really the, you know, that kind of hints that it's beneficial to have these, the connected corridor boost from the vehicle owners, but also at a system level.
Next slide:
So here's a – I'll show a video a little bit that show demonstration of that. So we kind of – we took like one point of the previous plot and just put it on a track. So with the experiment here as a real vehicle on a track in a single-lane corridor with virtual traffic signals and virtual preceding vehicles. And you'll see, so you'll have, so you have a vehicle that's automatically driven. This is work done in partnership with our colleagues at Clemson University. And OK, let me play it. And you'll see that as it drives, the automation system thinks it’s in this virtual world where there are traffic lights, there's a preceding vehicle. And you'll see it, one example of eco approach where the vehicle, the light is red so the preceding vehicle is stopped. Here we're slowing down but not stopping or coming close. And then we'll start moving back on when the light becomes green again. Just like right now. So we measure – actually we're able to measure the energy savings for that real vehicle in various situations, whereas there – it's an AV, automated vehicle, it has 15 percent energy savings. If it's in a connected automated vehicle, it has more energy savings. And also if it's a human driven, so here a human driven would still be through algorithm, that is a human model, we still get 50 percent savings if the traffic has CAVs present in a downstream traffic. So again in the spirit of the more you know, the more you get in terms of energy savings. so we've looked at automation with no good connectivity, to infrastructure how about connectivity, a vehicle to vehicle connectivity, V2V.
Next slide:
So here that can also bring energy savings. How would that work? So itself does not – it's just information, so you have to make sense of it. So let's, for example, so you wanna, you want to – for example, in this situation let's assume you receive information, basic safety messages from connected vehicles that are in downstream traffic. So you know their speed, their position. But for example, the preceding vehicle is not connected. So you only can sense it. But knowing what the downstream vehicles do can help you predict what your immediately preceding vehicle, the target here, what it will do. In that information, better knowledge leads to better control. And that leads to energy savings. So even at 30 percent energy savings, you can get – sorry, 30 percent penetration of connected vehicles, none are human-driven vehicles. So that they don't have to be automated here. That enables you to get 9 percent energy and 12 percent if you have the all 100 percent penetration. So definitely V2V technology could be considered for further energy-efficiency improvement. So we've seen how at the vehicle level we can – you know, automation with a connectivity or without it can save energy. We also need to look at the a bit more at the system level. And I'll pass on to Yunli for the next part of the presentation.
Yunli Shao:
Yep, thanks, Dominik. So I think we are talking a lot on the vehicle control side associated with the vehicle control so we can do an eco approach for example, and the vehicle acceleration deceleration will be everything, be smooth. But how about if we can add other components with a signal control logic, so if the signal light can also receive information from the vehicle, so that it can know what would be the, you know, vehicle demand and that you adjust the signal timing. So this is something I think we can integrate, to do a vehicle and signal control together, so that it can maximize the energy potential. In order to do this in … next slide, please.
Next slide:
So in order to do this, so to do the signal control algorithm, first thing is we want to study if we can do some advanced control algorithm to optimize the signal strategy. And to do the advanced control algorithm, first thing is you need to have a model. So if this is showing one of the example of the model is a bilinear control model, which is about the relationship between the traffic of traffic signal line timing versus the traffic delay. So with such a model, then we can describe the traffic dynamics and can optimize the overall delay and the volume of the corridor that we can improve the benefits. You can show that with such a bilinear controller we can have 9 percent reduction in the system-level traffic delay, compared to the real world, actually the signal control case. So here we compare to the real-world signal controller, so we get the rear signal time implant and the rear signal control logic and put it into the simulation and to compare the results. Because the – and here this is just showing up a linear control model as an example, but we can also use other kind of modeling of the traffic dynamics and then come up with other central strategy. But the point is, with this control strategy, we can not only minimize the delay but also since the delay is minimized, volume is increased. The traffic flow is much smoother than there is less stops and we also have 5 percent, actually, system-level energy savings, even though we are not directly optimizing and targeting the energy consumption. So this really shows that the power of advanced signal control algorithm, how that can improve both the mobility and the efficiency. Next slide, please.
Next slide:
With this signal control algorithm, now we can test how that can work together with the speed optimization algorithm. So here what we are comparing, the different metrics we can look at, you know, to understand how the system performs. Here the first thing we're looking at is comparing each individual vehicle. So the same vehicle travel at the same time and the same route from the optimal case versus a baseline case. Then we compare what would be the energy benefits and improvements for this 2D type of vehicle. Since we are comparing each individual vehicle, you can see there's some negative number, because the signal time is changed then the different vehicle they would enter maybe previously the baseline passes through with the green light. But now in this optimal case, it passes through with some stop, lock stops in the red light. So but the point really is the average sense in the sense that if the vehicle travels through the same corridor on a daily basis now on the average sense of how what would be the energy benefit for this vehicle. So that's what we are plotting here and trying to show here. So as we can see for different plottings really starting from as low as 20 percent of CAV that we can have I suppose the 22 percent energy savings compared to non-controlled vehicles. And this is really showing that the power of the integrated control strategy. Next slide, please.
Next slide:
So for the individual vehicle level, we see the benefit, but this is really from the vehicle perspective. But how about if there is a traffic management perspective, if the local DOT or maybe a policymaker, what do you see? OK, if there is a signal, integrated signal, and the vehicle control, how will the entire traffic performance be impacted? So if we look at how the vehicle’s traveling with this certain corridors we are studying, and factoring in the total energy consumption of all the traffic within this corridor from both the two directions, the eastbound and the westbound, now we can see the similar result is that as low as 20 percent we can already see quite a significant traffic energy savings. And the more [inaudible] rate we can have and we can have more benefits. And we can have up to 23 percent energy savings, you know, for the way, when there's everybody is connected vehicle. And specifically if you look at the plot on the right-hand side, this is showing the trajectory of each vehicle. And you can see the blue curves, those connected under those vehicles. They really can see previewing the, you know, the traffic signal light, and they can have a smoother pass through of the entire corridor to catch all the green windows. And they can really reduce the energy consumption. And because we are doing both the signal control and the vehicle speed control, so this can also help the vehicle to know the spot information better, since we know what would be the future spot information for the upcoming traffic light, especially for such a actuated control scenarios. So if – so essentially we have – we further validated the, you know, the impact and the benefits to have such a integrated control. And also we see that even though the low [inaudible] rate can have immediate benefits, we would wish, you know, higher benefit and a [inaudible] rate, which we can have more benefit. Since right now we don't have any connected – much connected vehicle and auto vehicle in a real-world road, so how do we really Implement that? That's something we would really talk about in the next few slides. If we go to next slide, please.
Next slide:
So first thing is we need to think about the impact of those human-driven vehicles. So how about the – how would they affect the behavior? And if there is a human vehicle and if there is queues in the intersections, then that would affect when those optimal vehicles or the CAVs need to stop at the intersection. So we studied the impact of the queue, prediction of the queue, to the energy savings. And we can see that with additional prediction on the queue at each intersection, there are additional energy savings that can be obtained. So this indicates that – so beyond the current standard video B or VQI or communication messages in the SA standard, if we can have additional information, if broadcast by the roadside unit, for example, for the queue information, then this can really be beneficial and help further increase the energy efficiency. Next slide, please.
Next slide:
Speaking of the real-world implementation side, another big concern is that for a lot of the video I devices, and if you have a dedicated communication unit in store at the signal light, and you need additional purchase, it's additional cost, and also need the installation, and also can cause some maintenance concerns. So we are really thinking about is there any way we can, you know, avoid, you know, doing all this kind of investment. Can we maybe say using a cloud-based solution for such eco driving scenario. And a lot of times, actually, the modern signal controllers, they’re already connected to a data center or maybe a traffic management center. So if the traffic management center can be the cloud here and all the vehicle, all the edge devices, they receive with the signal spark information from the cloud, then this can potentially avoid the investment on these equipments. So we did some laboratory tests first. So we see in different type of communication mechanism and how that will work. So from the 5G, which is directly through the mobile cellular to the data center or through the wireless communication, so like a wi-fi, or through a wired connection, so if we use really the ethernet cable, and we can see even though the wired connection has the best performance in terms of the latency and also the standard deviation, you know, it has the smallest standard deviation delays. The mobile hotspot 5G and the wireless connection is not too bad, especially considering that for a signal for the start information, the signal light won't change drastically from second to second. And the first actual application, I think the cloud, this indicates the cloud-based on the cellular based communication has a greater potential to address the information need. Go to next slide, please.
Next slide:
So previously – oops, go back one slide. So previously we are showing it in the laboratory so it's a perfect environment. So now we are also putting, developing a mobile app. So if you can help click that app, that should be a video. Thanks. So we developed a mobile app, which can be the speed guidance with all the speed control algorithm. And it can receive the spark information from the data center. So we can drive this vehicle around the real world and to see how the real-world performance looks like, what would be the delay, for example. As you can see, the delay increase comparing to the laboratory test, but it still – the delay’s well within the one second, which we see – we consider is sufficient for real-world implementation. And this will be adequate for such eco driving kind of application. So in safety critical applications, it would be more important for the delays, but for such energy, as long as we can know the signal light preview for the next, you know, few cycles. And even there is a one-second delay this will be tolerant for such application. Next slide, please.
Next slide:
So now I will switch topic a little bit. So we have been talking about the quite a bit different control applications and the different potential of the energy savings for different algorithms. So another big chunk is that how do we even evaluate this kind of control algorithm? It's quite a complex system if you think about it. So even for each connection to the vehicle, it involves system of systems and with many different components from the localization perception, past planning, or maybe control algorithm to the vehicle dynamics. So and this – it's not unlike, you know, traditionally, so when we look at the vehicle we can isolate it with just one eco vehicle. Now since we need to do the communication, we need to interact with all the surrounding vehicles. We also need to consider the traffic perspective. So we require some kind of a traffic simulation to study to have a realistic traffic scenario and the traffic environment beyond the vehicle dynamics or the sensors or the powertrain dynamics. And this requires a co-simulation environment of both the vehicle simulator and the traffic simulator. And then in addition, as Dominik mentioned, that we – at both labs we are starting XIL framework. So if we have hardware in the loop, then we can have the actual system dynamics and the responses, which can otherwise be complex to model and measure. So with all these kind of components put in together, we can provide a high fidelity or scenario within an XIL or co-simulation. This can really shift a lot of on-road testing efforts to a laboratory and high-fidelity virtual environment to accelerate the curve validation. Next slide, please.
Next slide:
As there are many components we need to put in together and the talk and the integrator together, so one question is how do we even make sure everybody talks to – all the components talk to each other? And how do we ensure all the software and hardware they are synchronized? And a lot of times the different software, they have different – they are written in different languages and a different algorithm. They may be written in different programming languages. So this really requires a flexible and low latency protocol to integrate and co-simulate these different components. One of the candidates can support such application is using the network internet communication protocols, such as the TCP. They will originally designed for such massive communication traffic, and they can have pretty flexible connection as long as the hardware is, of course, are using the cable connection. And also it can have very low latency to support a large scale, potentially a larger-scale scenario to study. Maybe there is a more penetration rate over the connected V and autonomous vehicles. Next slide, please.
Next slide:
With the bankable of the code simulation established, then the next question is how do we even ensure this virtual environment that can reflect in the real world? So to create a real-world scenario in this virtual environment, it really can make sure we can test all the different country algorithms. I mean, there's a virtual word first, with a very repeatable, very controllable environment. And we can shift a lot of our efforts in the virtual environment. And also we can avoid a lot of the safety concerns. And this requires not only just the vehicle data from the unloaded vehicle connection. And also this requires the traffic data may be collected from the roadside unit or maybe from the traffic signal light. And also that they require the GIS data to define the geometry and specific network we want to study. And we are trying to develop a flexible unloaded data connection system. So as you can see, we have instrumented vehicle without the lighter and the cameras and then we can collect the Android vehicle environments. And also with the traffic data we're trying to collaborate and connecting with the local DOT, if you collect all these traffic data. For these different – the challenging part of this is that for different data sources and different applications they have a different requirements. So if we are most thinking about the vehicle perspective, then this can be the sales report and all the 3D digital twin part. So how real is real? How many details we need for these different digital twins? Then for the traffic side, if we're doing signal controller for example. So how do we create a realistic traffic signal light and the traffic signal scenario? So here already different developers or different researchers typically focusing on what specific traffic network. And it's challenging to transfer and share this kind of digital twin scenarios. So we really find the need to standardize our different scenarios. So that we can have common bases so that we can share with among different researchers for vehicle development and validation. Next slide, please.
Next slide:
And also we want – another thing I want to touch base is that we want to verify a little bit on the dynamics or the XIL approaches. So we talk about the XIL and the need for very detailed physicals into this whole code simulation environment. So how critical it is. This is really showing a case study. So let's say if both the passenger electric vehicle and the class is a truck. If we are given the same control optimal control command to these two vehicles, and considering very detailed dynamics of them, they would naturally have different responses. As you can see, since the Class 8 truck they have a slower dynamics, especially they have all the transmissions as you can see here when they are doing – when the vehicle is doing the accelerating. So they cannot really imagine the control command due to all this dynamics we need to consider when we're designing the control. And even for the passenger if it has a faster response. And when it's decelerating, as you can see there is a very small – it's a very small difference, it looks like a control command an actual responses. But even this is small risk differences in the speed, it can cause quite a significant difference in the following distance, which can be critical from a safety perspective or maybe from a low demand shift perspective. So really this really shows the need that you have detailed vehicle dynamics when evaluating CAV control, especially for heavy-duty vehicles. Next slide, please.
Next slide:
So as a summary or a recap of what we have covered here today. So there's many current state of the art. And we are seeing that for this current state of art, there's a future potential areas we can focus. So first is that so currently real-world sensor performance are not really captured with nominal specifications. Second is that limited information such as queue length and signal light limits the green light speed advisory. Then the production automated driving systems focused on mainly on the safety side, but they have a greater – they can lead to a potential energy penalties, since we need to trade off between the safety and the energy. And also, when we’re considering a lot of scenarios like cut-ins or lane changes, that can lead to energy consumption increase. So as a future focus or a potential of area to facilitate the implementation or the future research, I think the representative scenario and the human driver are critical to estimate the CAVs’ energy impact. And we can see a 25 percent energy savings demonstrated through V2I communications. And the more we can know, the more we can save. So if we can know more about the powertrain dynamics, then we can develop a specific control for the powertrain. And then we can consider multiple traffic lights, but you can consider V2V configuration in addition to the V2I. And also, if we can share a new addition to just kind of the speed information or location information, we can share the intent of the vehicle as additional information. Or we can integrate the traffic light and the vehicle control. I think, and also I want to add that we have to see that, you know, for the cloud-based service, it can be potentially another alternative for V2I communication. And also for the XIL, code simulation can help shift a lot of the testing experiments to a high-fidelity virtual environment to accelerate that development. With that, I would turn back into Kevin.
Kevin Viita:
Next slide:
Thank-you, Yunli. Before we go into the Q&A portion, real quick I'd like to ask the audience to follow the link in the chat regarding the post webinar feedback survey. If you could fill that out after today's webinar, that would help us improve our webinars going forward through the series and for future webinars we host with the Department of Energy. Now, a couple questions real quick: How does electric versus non-electric vehicles affect these findings? Is it more important for non-electric vehicles than electric?
Yunli Shao:
So I can take this question, Kevin. Thank-you for the question. So, and there's another question about EV range. Generally what we've seen is that we have comparable energy savings for EVs and conventional vehicles, so it could really benefit both technologies. And yeah, and you know, if we turn that into range, you know, 20 percent – for example, if we take a 20 percent energy saving for a EV that has let's say 400 miles electric range, that would add 100 more miles to its range. So there's a opportunity to increase the range through for these technologies.
Kevin Viita:
Fantastic. Thinking about your example of the eco approach regarding V2X, have you looked at user acceptance? I can imagine that the passenger will have expectations about speed and spacing between vehicles.
Yunli Shao:
Yeah, I can take that one. Yeah, I think that's a that's an excellent question. That's something we are actively investigating in our current setup, since mostly we are doing a simulation or hardware in loop studies. So we are considering to reduce the impact of the vehicle behind it to include some of those terms, like you know, not to be too far away from the speed limit. Even we are kind of gliding to a single light, we don't want you to, you know, to be too low comparing to the speed limit and the surrounding vehicles. Those are already considered. But as you further evaluate that, that's some of the current activities we're trying to do, since we are doing – for example, we have some laboratory, at the Oak Ridge National Lab. So we have the vehicle on the dyno in a vehicle in the loop setup and big screens. So potentially we can do some human studies to see, OK, so how would if – let's say the big screen is showing an eco vehicle. You know, is doing optimization and gliding in front of that, how this human be human to perceive that, and you react to that. This is how some of the current studies we want to understand and to clearly know better, and also I think as we are thinking about a more app based application. So if we can have more apps deployed in real life and more people are using that, potentially people would have more anticipation about how, you know, basically if the driver, surrounding driver will drive differently than today. So I think this would also gradually help in, you know, people to perceive and to anticipate and can reduce some of the issues.
Kevin Viita:
OK, my apologies, I know there were a lot of questions that came in. But I think we have time for one more here. And it's regarding when you're talking about CAVs, what do you talk – what are you thinking in terms of how connected will they be? Assuming CAV vehicles will run with platoons, are we talking just vehicle to infrastructure communications, like SPaT, or will there be platooning, and would that lead to further energy savings?
Dominik Karbowski:
So connectivity can be through like CV2X technology, and now that the SRC is kind of gone, or through the cloud. And those are really the main key technologies. So really, we looked more at SPaT broadcasting in their, you know, that's – that can be achieved through cloud, for example. And V2V is one where there will be basic safety messages and things like that. But so, you know, of course, these are – will be more – there will be more challenges putting these on the road. You know, so we're – I think in SMART, we're trying to show what is the potential if more of these technology would be turned into real world.
Kevin Viita:
Well, thank-you, Dominik, and thank-you, Yunli, for the presentations today. And thank-you for answering a few of the questions that came in. I want to thank you guys for presenting. Like I mentioned at the beginning, the slides and the recorded version of this webinar will be available from the EEMS website. I believe I included a link in the – I did not. They will be available on the EEMS website. Thank-you, everyone, for coming, and our next webinar will be Tuesday, June 20. We'll see you then. Thank-you for coming.