COVID-19: Where are the self-driving cars and trucks?
5:47PM Apr 24, 2020
Morning, Good afternoon. Good evening, and welcome to episode five of the AI for Good webinar series. We hope that you your family, your friends and your colleagues are all safe and keeping healthy. My name is Fred Werner from the ITU, the International Telecommunication Union. And it's my privilege to introduce today's webinar episode.
So who is the ITU? Well, basically your mobile phone would not work without vi to you because the ITU is responsible for allocating and coordinating radio spectrum. It also develops technical standards that underpin modern telecommunication networks.
For example, you're looking at this video, there's a very good chance it will developed using video compression technology that was standardized by to you alongside ISO and IEC. It will also assist developing countries and bringing up their ICT infrastructure through capacity building, training and development work.
Last but not least, the ITU is responsible for organizing the AI for Good Global Summit, alongside XPrize foundation in partnership with 36 UN agencies, the ACM, and Switzerland. And like much of world, the AI for Good Global Summit had just decided to go digital. And we're now featuring weekly programming online throughout the year. So you could consider this episode to be part one of the smart and safe mobility track that would have taken place at the physical summit. But now we're offering it digitally.
And the title of today's episode is COVID-19. Where are all the self driving cars and trucks? And that's a very good question and we're very fortunate to have with us here today. panelists, a panel of AI and autonomous driving experts that will help us to learn more about this subject.
But first, some housekeeping rules. We have disabled your microphones, so please use the chat and the Q&A function if you wish to communicate. The moderator will be responsible for identifying and answering questions that have been posed through that system. And we're counting on your active participation to create a very interactive session here today. So without ado, I'd like to introduce the moderator of today's session, Mr. Bryn Balcombe. He's the CSO of Roborace and ADA, the autonomous driving Alliance. He's also the chair of the smart and safe mobility track from the AI for Good Global Summit. And he's the chair of the ITU Focus Group on AI and autonomous and Assisted Driving. So Bryn I'd like to hand the floor over to you and the show is all yours. Welcome.
Thank thank you very much for it's an honor to be presenting. I've followed the AI for Good summit now for the last year. For years, it's a great initiative. So to see this coming line is, is fantastic. And it's a great opportunity. So can you hear me okay? So for you just pause your video freaked me out.
So today I'm joined by Michelle Avary, head of automotive and autonomous mobility at the World Economic Forum. Hello Michelle, if you'd like to introduce yourself a little bit more, it'd be great.
Yes, hello, thank you. And we appreciate the opportunity to be here.
The World Economic Forum is a private public partnership whose mission is to improve the state of the world. And when this relates to mobility, we work on three predominant areas to make sure mobility is safe, that it's clean and that it's also inclusive. And we work at the intersection where the government is unable to solve the problem. An industry can't solve the problem unless we come together to solve those problems. So we work in the intersection of policy, regulation, and industry collaboration.
Thank you very much, Michelle. And then I'm also joined by Misty Cummings, Missy Cummings, who's professor at Duke University, and notably one of the US Navy's first female fighter pilots. So welcome, Missy. give some background as well, that would be fantastic.
Thanks for having me. And so that's right. I'm a professor at Duke University. I'm in Duke robotics and the lab that I direct is the humans and autonomy laboratory. And my research really looks at the intersection of humans and autonomous technology, and relevant to driverless cars. I have basically two thrusts. The first is looking at humans who are operating in and around, autonomous system. So I do a lot of work in driverless cars and drones. But we also look at how humans design these technologies. I think it comes as a big shock to a lot of people that autonomous technologies, artificial intelligence, machine learning, deep learning, these are actually fairly subjective, mathematical approaches to solving a problem. And so it turns out that there's a lot of human bias that can kind of creep into these technologies. And so we're also looking at better ways to design autonomous technologies.
Super. Great. Thank you very much. So I'm just going to give a give everyone a brief opening just a bit of background before we start the main conversation. So I wrote this a little bit earlier today as a summary, so it's late April, we're four months into the COVID-19 pandemic, to date the WHO reported 25 million cases globally, and 175,000 fatalities to slow the right of transmission. governments around the world have implemented travel restrictions, quarantines curfews Stay-at-home orders and facility closures for all but essential workers. COVID-19 is undoubtedly a global health emergency. Yet the situation that simultaneous global response has created the world's largest social experiment. Research from the University College London show that it takes between three weeks and eight months to change your habit. And on average, just two months. So how much of the behavioral change we're experiencing will persist after the pandemic? What will miss what what will this shape and shape our new normal? How will that look like coincidentally in 2020, was positioned as a pivotal year to address global warming. In January, I saw at the UN Climate Change Deputy Executive Secretary stated the oceans are acidifying. The soil is degrading crops are becoming less nutritious. desertification is spreading the ice caps are melting and we are destroying biodiversity. Climate change is not slowing down. We're not acting with enough urgency to address it. So in many ways, to me, it feels like Mother Nature has slapped us all around the face to take the necessary collective action. So COVID-19 to the tragedy for so many people, yet it's also our greatest opportunity to refocus on what is truly essential for our future. So with that said, let's kick things off with some questions. So, Michelle, Missy, and 2020, was billed as Prime Time for the self driving car industry. So where are the self driving cars and trucks we were promised and Why are they here? So Michelle, if you'd like to kick off that would be great.
Sure, well, obviously here in the United States and in the San Francisco Bay area where I'm located. Most of the all of the self driving vehicle, pilots and testing has been stopped and we're honoring all the shelter in place orders. In places like China where we also have an office and we have people working there. They're back on the roads. And in fact, in China we saw many autonomous vehicles particularly in the delivery logistics space, really stepping in to fill in that gap that we have on the transportation capacity. So we saw many a V's providing non contact, low and medium speed products, most notably JD COMM The Baidu Apollo Neo Lex, and they were really addressing those transportation capacity constraints. Obviously when you have road closures and lack of traffic, it made bringing these vehicles out and widespread or wider spread public usage a lot more palatable.
Michelle, do you think it's a regulation or technology that's kind of made that we haven't seen the dawn of autonomous vehicles for where we are now? Because the promise was that we would see these things on the road around 2020 numbers varies, but it was around 2020. So, you know,
well, I think it's three things. I think we need clarity on regulation globally, without a doubt that's needed. But we also need technological validity, as well as business model validity. One thing COVID-19 seems to be showing is there actually is a potentially very lucrative business model within this logistics and delivery space. for low and medium speed products. We'll see if the the learning and the data that we're getting is applicable when we start looking at higher speed, highly automated driving systems and how that looks but we do need a clear path to market We need regulator of regulation, clarity, as well as technical validity.
And Missy from, from your perspective, why would you say we haven't seen the dawn of self driving cars or trucks at this moment in time?
Well, I hate to be the nerd in the room. But there's actually a very specific reason why they haven't come to fruition. And that's really the problem with their perception systems. And so what that means is how the cars see the world. Most people believe that driverless cars need a combination of camera vision, LIDAR radar, you know, some manufacturers notably, Tesla thinks that they can do it with just camera vision, and radar. So but even when you have when you add LIDAR to the mix, there just some limitations about when the sensors can see the world in the way that the world needs to be seen under many different conditions, right? Snow, for example. And so this is actually why the speed that we just heard about speed does matter. So at highway speeds, the perception systems have to compute that much faster and take in that much more information. So that's why we're seeing cars on the highway. And this would also include things like platooning trucks have a much more difficult problem. Whereas slower speed vehicles, slow speed delivery shuttles, maybe some slow speed, multi passenger transport is having a better time at perceiving the world. I want to just qualify that to say that while slow speed does help, it still hasn't resolved all the issues around fragility of perception systems.
Yeah, okay. It's, it's interesting, because after all you get to after you've solved perception, there's still a problem with prediction in terms of predicting the behavior of all the other road users and I think that's the That's actually something that will be coming to the fore as we progress in the perception quality system. But missing one at one of the other things, you know, I was when I read your paper on artificial intelligence and the future of warfare, so you obviously still have some contacts in that domain. But you make a very interesting distinction between automated and autonomous which kind of sits somewhat against what the SAE say. So be interesting to, to explore that a little bit. Yeah. What would you say is the difference between automated and autonomous?
Yeah, so I think it's an important distinction that applies across a lot of communities, not just driving. But automated systems are systems that operate by deterministic rules. So when an algorithm runs in an automated system, it runs the same way every time and thus you have a very high certainty of how the system is going to perform. And autonomous system is making guesses about the world around it. And because of that, it embeds probabilistic reasoning algorithms. It's these stochastic algorithms. That's a very fancy Professor word uncertainty, but because of the high uncertainty, then that is why the system's autonomous systems are having to make the most reasonable guess under the circumstances that it can make. And so understanding that some systems can be very successful being automated. So robots in car manufacturing facilities, for example, are highly automated. They operate the same way every time very precise, very repeatable. One of the problems with driverless cars and this is becoming particularly acute in certifying them and this is true of military autonomous systems, too, is that when the system doesn't perform the same way every time, even under the most benign conditions, then how can we ever put guarantees on its ability to navigate the world safely?
right now. It's like how we deal with teenagers. You know, When we think about it as No, seriously, when you think about it, I do have a teenage son. So I think a good bit about this. But when you think about how we get a learner's permit or driver's license that actually doesn't assess how safe I'm going to be on the road. Um, we and I know, Brian, you've been very involved in some of the work we've been doing at the forum, to really look at these issues using a scenario based approach to AV safety. And we think this is a key way that we can help regulators truly understand what the safety is of highly automated driving system in the context of its environment. And that's based on how we think the vehicle should respond safely to factors in the environment, whether it's pedestrians, intersections, cyclists, garbage trucks, and then really build a better framework. That is Forgive me operational design domain specific.
Yeah, it's very, it's, it's fascinating actually, because, you know, when I look back, I pass my test at 17. I think I did it in 17 hours of learning. But the moment I've done that, in the UK, I'm then able to travel internationally and drive internationally, you know, come over to the US and drive on the opposite side of the road in vehicles that are incredibly large with road regulations. I have no idea what they are not giving any guidance as to what they are. And yet I'm still able to, to transfer if you like my knowledge and my learning from one environment to another and drive safely. So I think that's a really interesting. We're not there yet with AI autonomous systems. And what we can see so far,
well, in some will say we're not there with people yet either that we have an operator problem. The FAA reports that 3500 people are killed on our roads. Every single day, and even though we see a huge reduction in the amount of traffic and people on the roads right now, the amount of speeding hasn't declined. And the fact that there are still fatalities and collisions happening, even in here in the Bay Area is frankly quite shocking. Because while I think we can agree, there's never a good time to get into an auto collision, now is a really bad time to be entering into an emergency room. But if you think about that sheer number of 1.3 million deaths a year and 50 million injuries, clearly we have an operator problem when it comes to like passenger vehicles or road vehicles. And that's where I think this this promise of highly automated driving systems, really it is a promise I understand that but I think we really need to go forward and look at in this idea, this ideal world what is the new approach To mobility safety, and it shouldn't just focus on reducing fatalities caused by vehicles, but also as your in your opening comments on the vehicles polluting effects, which we know are very detrimental. So we really support going with alternative power trains. But we also believe that we need to have a safe use of the shared public spaces. So it can be safer for pedestrians, cyclists, motorcyclists, as well as protecting the public health by eliminating these opportunities for contagion, which was really part of that biosafety view of what we saw coming out of China in using these non contact low speed, highly automated vehicles for delivery in areas that were locked down. We really should explore more of these and think about where we can remove some of the biosafety risks to our essential workers. Can we minimize
Yeah, for sure, Missy when you when we talk about automated and autonomous, but you also have an interesting graduation scheme, I think which goes from skills to rules to knowledge to expertise. And it's a very interesting path, I think would be good to take the listeners through that, if that's possible.
Yeah. So I would, of course encourage you to go to my website at Duke University, just type my name into Google and it will get you there. And I have a whole series of papers that show you how autonomous systems reason and how humans reason. And it's one of the things that we're seeing quite clearly is, as long as uncertainty is low and the speed of time that the speed that systems can reason and is low enough, that autonomous systems really prevail and something we call skill and rule based reasoning. So this is why most planes are highly automated because they don't need to have fancy perception autonomous perception sensors planes are automated planes that take off and land by themselves are have a high degree of automation. And under low uncertainty they do well the problem where we start to break that down is when something goes wrong in areas of high uncertainty and so the example I like to give is Chesley Sullenberger and the miracle of the Hudson right so the plane to engine failure valve, the commercial plane as a glider, a decision had to be made maximum uncertainty, could the plane get to a runway was it going to have to be land to be put down in the water. And so he ended up being able to cope with massive uncertainty with extremely good outcomes that doubtful would have happened had an automated or autonomous system been in place. Now that's not to say that we never will get there. We're just not there right now. And so I'm a big fan of all sorts of Automated and autonomous systems. As long as we understand something that I call function allocation, do you really understand what functions should be done by a computer? And do you understand what functions should be done by human? Sometimes they can be different. You know, in the case of automated landing, airplanes do better all the time when I was a carrier pilot, the automation always did a better landing as a carrier pilot. But under when there's an emergency, and there's something wrong with the automation, then there's a problem. And in fact, we saw that humans can break down under these circumstances as well and the Asiana crash in the bay area where the pilots were too used to having so much automation that they had skill degradation. And so in that particular case, the airlines did not understand it because of the degree of automation. But the fact that the automation wasn't perfect, and it would have to be taken down. Every now and then that we still need a human. And we still need to train the human. So I this is going to be different for every domain that you apply it to. But if you don't understand when your system breaks down under uncertainty and you don't have a human, a correctly trained human to intervene under those times of uncertainty, then you're going to have traumatic outcomes. But that being said, You know, I do believe that moving into a an autonomous driving scenario, I think the best balance that you're going to get there is under slow speed operations. And in the short term, I think slow speed, delivery of goods, whether it's groceries, for example, or medicine, I think that that is the driverless car domain that has the best chance of surviving and I do mean that I think that the COVID-19 scenario is going to really decimate The driverless car community just before this happens Starsky, the automated freight company went under, I think you're going to see more driverless car technology start to go under those companies that survive will be the ones that figure out that the delivery slow speed operations is the right way to go in the near term.
Yeah, no, I would agree. And it's interesting to see the parallels I think what we'll we're learning is that autonomous and robotic systems apply doesn't matter what the domains you were talking about airlines, it applies, some of the same philosophies applies. So that over trust of current level two systems that we're seeing on the road is actually a big issue. It's a big safety issue. And when it comes to training, untrained operators, they're not given any guidance as to where or when these systems might fail. They're just given the executive order to say, You are responsible, I'm sorry, you are responsible is that you are the driver of this vehicle. We're assisting You You are responsible now, right? That that is really difficult to then understand. But when is this system going to fail. And I tend to equate it is like the, the hype has always been about, okay, we can sit in our cars, read our books, and relax, it takes away the load. But the reality of a level two system is that you act like a driving instructor, you have someone else who's driving the car for you. And it's your job to monitor not only the environment and the risk that's presented by the environment, but whether the other driver is perceiving that environment correctly and taking the appropriate actions. And if it isn't, it's your job to intervene.
But we're already we're in we're doing that already with our cars and with our advanced driver assistance systems. We're already confusing vehicle owners and operators. And if you look out every single day, you have anecdotal evidence that people actually don't like to drive. I hear it all. The time that I love driving, I'm never gonna give it up. And I really challenge that because we see so many people on the roads today doing anything other than actually effectively controlling that 3000 pound vehicle that can and does kill people. We I mean, it's that we know this enough. So it's what can we do to help streamline a das systems help really better train the operator and I admit I am one. I am the reason why we want highly automated driving systems. I'm terrible on the road. I drive very aggressively. I feel like I own the place. And I really feel that a lot of people are really terrible operators, even with Ada systems as well, and that we can do something better. And when we think about the shakeout, that's happening in the Silicon Valley and also with within the highly automated driving space, there is a flip side to it, you know, there is still venture capital out there, they're still sitting on a lot of money and they need to allocate those monies. Or they're not even they don't even have a chance of getting any return on it. So there is still money in the in this industry. And it also means is, as some of these companies start struggling, that the cost to invest in these companies has just decreased. So I'm actually hoping that as the shakeout happens, that those really great startups start rising to the top and maybe do get acquired or we do see some more consolidation happening with them. Because I think it would be really unfortunate if we didn't press forward in developing highly automated driving systems for mass scale, even if it is beginning in the delivery space and I agree completely with Missy, that this is where we need to be Pointing our interest rate now in our efforts.
Missy, obviously, a lot of the self driving car industry started with a DARPA challenge. They started sort of in the military domain, and it's progressed into the, you know, the civil domain, if you like the commercial domain. And I think you've made some interesting comments before about the, the, the, as Michelle was mentioning the cost of developing this technology and the price point for the talent because the talent is so rare, having an impact on military development, so it'd be good to get your insights on that and how you see that changing?
Yeah, so I think that that what has happened in the last 10 years is probably the strangest inversion of talent, attraction that we've ever seen. The military did actually develop all this autonomous vehicle technology, and then it jumped into the commercial sphere and totally skipped military development, which is normally what would happen and so now What has happened in the country is that all the real top talent is either in Silicon Valley or wants to be in Silicon Valley and working on commercial technologies. You know, which is great, except that now what we have is, I call it the tumbleweed going across defense industry because the defense industry in the United States at least and I suspect, this is true in some countries globally, that we don't have enough top talent going into defense industry. And so the autonomous vehicle development is happening way faster in the commercial world and the military world. I tell people all the time that it is very possible that there could be a some kind of conflict situation where civilians have access to flying cars, and military people do not. And when we have that kind of weird technology inversion where there's more capable technology in the city billion world and in the military world, there's obviously some defense and implications that aren't good. And we need to really rethink then how, you know, how are we attracting talent? And and I think the defense industry really needs to look hard internally to change the way they're doing business. Right. So people don't want to work in the defense world. So I think the defense industry and to a large part academia also needs to think about how they're training people. Yeah,
no, I think it's gonna be a really interesting, interesting shift that comes from that. I think we said before I am robotics as as generic fields look like they're going to continue to increase over time, you know, and I think COVID-19 has even shown that in various different sectors. So the expertise that's being built will continue. It will then be interesting to see which markets that's going to be dominant in. You know, I think we've already said with the IV market, we may see a shift From the carrying of people to carrying of goods that, Michelle, we've discussed before the carrying of people as a premium that the carrying carrying of goods doesn't at the moment so commercially is more attractive to move people and goods.
Yeah, the margins do tend to be higher. And I think we're seeing that right now in the ride hailing industry, where a drivers ability to make money is better when they move people than when they're delivering food. It just looked a lot the economics of it makes sense. I do think if we look at the future scenario of this, contact lists or low contact and low or medium speed products as being ideal for highly automated driving systems. We should look at public transit. And look at where we know. Well. Companies like transdev Have an operate fixed route systems. We see here in the Bay Area, we took our bus routes down from 9090 routes to about 17 routes. And in a city of 700,000 people, we still have 100,000 people who are central workers who need to get to work. The subways are closed, the light rails are closed, these people are really, really stuck. And what happens is you have a lot of stress on the drivers of these buses and the operators of these systems and the cleaners have these systems. And I think it's worth when as we look forward thinking about biosafety is at what point are we putting drivers at risk as well? And is there an opportunity to do a couple of things. One is look at not only making sure central workers are safe, by taking some risks out of the system by bringing in highly optimal It's driving systems. But also can we look at reducing some of the future cost of public transit? Because we know that for public transit right now, it they're usually subsidized by about $1 per ride. But we know that ridership is going to go down probably orders of magnitude of 50, or 60%. We're seeing that in China right now where they're back open, but public transit has not picked up, it's picked up at about 30% compared to traffic is back at 70 or 80% of normal. So if we look at that we need to really balanced this need for getting people back to work securing important jobs of which bus drivers and transit operators are, but also balancing the biosafety risks as well as the cost. And there's no easy answer to this strategic tensions that exist. The only way to move forward with this is to have community based discussions on how we can deploy some of these systems where we can, and how we should do it to not only protect workers in jobs, but also biosafety as well.
You know, missing one of the latest papers that I did manage to finish reading yesterday was about autonomous vehicle dispatch operations. So you know, vnl be unable to then control all of these entities in the environment. And I think that's also important. It's something that's often overlooked, but you've made some good comments about way mo in their facilities. But managing these vehicles and managing them safely. managing them remotely, becomes really important when we're looking at mobility and shared mobility in particular.
Yes, so I think this is an often overlooked area about what does remote control or remote operation of these vehicles look like. And so I've recently released a report on my website that addresses all of those factors. So one of the questions I get asked quite a bit is why can't we just remember remotely control cars. If they if they can't operate all the time in all conditions, then why can't we just have a building somewhere in the world and have people ready, almost like a driving game where they made their way out of it and you know, for the longest time, and certainly that's how we've been operating drones in the early days of drones, they were remotely flown that way. But one of the things that we found out in the early days of drone operations was that when you had somebody do the takeoff and landing remotely, that there was a high crash rate and eventually the Air Force, US Air Force had to ban people from remotely operating takeoffs and landings that had to be purely automated. And the reason was because of the time latency in the control loops. And these issues also are very relevant for driverless cars. There is a time delay that is going to be that occurs in any kind of transmission and then there's something else called the neuromuscular lag. If it's inherent in every human that, that from the time you're I see something and it can actually transition in your head to an action with your hand, then, you know, you're you're looking at about a half second delay. And for you know, that's quick, but a lot of bad things can happen in the driving domain, especially at highway speeds and a half second. And so if you're expecting a remote control driver who has to perceive the world with their half second neuromuscular lag, and then the lag of the communications network, and both ways, you're really just setting up this situation for very, very bad outcomes. So I say no, no, no, no remote operation. driverless cars, unless, unless it's like very slow things like backing a car out of a ditch or something like that. But then there's a whole nother layer of things we have to think about. How are we gonna put Have these dispatch services? How many people do you need to supervise a fleet of cars? The reason Wei mo had to shut down in the COVID-19 crisis, they had to shut down their purely driverless car fleet and in Arizona, and people want to know, why is that if the cars are driving themselves, why do we need to shut that down? I would say number one, there's a sanitation issue inside the car, you know, if you're going to change out passengers, that's a nightmare for how do you guarantee cleanliness in a car. But it also turns out that there are so many people that are needed right now to supervise driverless cars inside these remote operation centers that you can there were too many people inside the remote operation centers. And so this kind of points to a larger problem that we have with autonomous systems that people don't often realize is that while some aspects of human tasking disappear, often it moves the responsibility of humans to some other place. And in this case of driverless car remote supervision, at least with current technology, the remote operation centers are so populated, that it's not clear yet that we've reached a point where the business model really makes sense. Yeah, we
we've seen this in telematics for 20 years that we've we've had connected vehicles that can respond to the vision in the event, excuse me, of, of an airbag deployment, you've got automatic airbag notification, where you have emergency personnel come on and help people find out where the vehicle is and the safety of the occupants and then deliver emergency services to them. So we're really very, very familiar with what it takes to using cellular systems and GPS systems to monitor vehicles remotely and in some instances control them remotely and there are limitations to it. And as we look at highly automated driving systems, particularly in a fleet environment, there's no doubt that remote monitoring is going to be necessary for this, not just for dealing with the cleanliness factor, but also dealing with occupant safety. If you look at public transit systems right now, a place like New York City, probably a third or so of the police force is dedicated to policing transit systems. Because we know that if we expand our definition of safety from let's not just run over people or hit things, let's not just worry about biosafety, but you actually do have to worry about your physical safety when you're unfortunately with other people because assaults, murderers these things happen. They happen to DD drivers, they happen in the systems as well. So there will still need to This idea of monitor when it comes to operations, I think there are some scenarios where it can be possible to help remotely move a vehicle in these low speed situations, but not fully operational as you would as you would if it was me.
Yeah, missy. And you said before there were almost two types of remote operations. So you have this sort of steering wheel throttle brake, you know, you you remove those and you're trying to do real time comms. But there's also some more goal directed ways of controlling these vehicles remotely, which may have some safety benefits or some other some other benefits actually, in terms of latencies. So,
yes, right. So it's a difference. I'm going to put my nerd hat on again, there's a difference between rate control, remote control and gold based, remote control. And indeed, almost every and every system I've ever been in It's never a good idea to give anyone rate control remotely because of those time latencies that we're talking about. So if we took out the rate control from drone pilots, how do they operate now? Well, they actually just put waypoints into the system and tell the aircraft where to go, and then the aircraft locally figures out where to go. So that is going to be the key to the future of fleets of automated vehicles, is that they can be commanded with goals, and then locally, the cars can figure out where they need to go to execute the goals. So there are some distinct advantages that we will be able to when we have fleets of cars that we can command in this way, even at low market penetration rates. There could be some very profound impacts. One of the problems that we have in North Carolina, for example, are hurricanes. And when a hurricane is coming towards the coast, we have to reverse all the lanes, so on All, you know, six lanes on an interstate will be reversed from two way to one way out traffic. And if you have remote vehicles that you can command by two goals to say basically either turn around or just don't ever go east always go west, then you can actually get evacuations to be much safer, much more efficient, much less congested. And so the idea of the commanding fleets of vehicles in emergencies or you can even imagine, after a big game, any kind of sports game that cars could help influence traffic, even with low market penetration rates, and there have been a lot of simulations that can show some real advantages. So I'm a big fan of thinking through this. I think one of the most important things people need to realize is even though the technology isn't here today, for you to buy your own driverless car and be able to jump in the backseat go to Las Vegas, for example. I'm not saying it never will be. It's just not here today and we don't have the sensor technology. I see a lot of questions in the q&a about, well, what is the right set of sensors? When are we going to have the right perception systems? The answer is that's why we still need to give money to universities to research. Of course, I'm going to university I'm going to say that but there's just some fundamental challenges, where we need to figure out some sensor limitations and some processing limitations. I am very optimistic about the long term future. I think in the short term, we can get driverless delivery vehicles, slow speed, but in the long term has got a limit.
Well, I also care with the shuttles with the low speed shuttles, which we see in operation in many places, and there are companies out there that are doing remote monitoring and remote assistance. There's a combinate called designate Driver that is working on these things on low speed shuttles quite effectively. And I think we need to continue not only investing in these types of startups in these companies, but also continue deploying in them and getting real data. So once we get out of these low speed, highly constrained, geo fenced situations, we can grow into these broader higher speed movement of people more completely.
But I find it fascinating actually back in there in the UK, there used to be a speed limit when the when the automobile first came out when people didn't trust it over the horse that was about 20 miles an hour. And 20 miles an hour now seems to be the number that we're all heading back towards because actually, there's less chance of a fatal collision at 20 miles an hour. So if you're in an urban environment, that's really the target speed that cities are now going for. So when we talk about location lead over mobility, the 20 mile an hour mark seems to be about the right target. So you've severely reduce the level of risk that's associated with operating that service in the first place. So that can compensate for the sort of where the technology is at the moment.
Yeah, we just want to make sure that in those situations that we're sharing enough of the data, and the learnings, whether it's on campuses, university campuses, on shuttles, whether it's in court facilities, we really need to make sure that we're sharing these the data in the learning more widely because we don't think that every single company and every single operator needs to learn safety firsthand. We think that if there is an incident, we should really be learning from each other on these, particularly for these outliers. And that's one of the things that we had the forum are really looking for at doing with the safety pool project, and with the work we're doing on data for AV safety is to begin sharing these learnings because some of these really are not competitive advantages. There's plenty of places to compete on, but we don't believe that core safety is one of them.
Yeah. And I think, basically, when you start to look at the dispatch centers, at the moment, there's a metric within California, which tends to look at disengagement as kind of an indicator as to how well the system is performing or not performing. You know, so what what do you think those metrics will be in the future? As we move to the dispatch centers? What What do you imagine they're going to be monitoring? You know, you've mentioned things like having a very accurate world model viewpoint before so making sure that vehicles are aware of the situations and the risks that are exhibited by the situations. So do you think it'll be those types of metrics that will come to the fore in the future?
Well, you know, I'm First of all, I I just want to reiterate For those of you who know me, your this is going to sound like a broken record, just engagements are not a good metric. And so disengagement only show you a partial picture and I've been strongly advocating both on Capitol Hill and more broadly across the community and, and I'm, I'm reaching out to the World Economic Forum, please let's work together on helping establish what I call a vision test for cars. Since we know that the perception systems right now are the long pole in the tent. I think that we really as a community, and I don't want to leave this up to the regulatory agencies because the National Highway Traffic Safety Administration is completely inept in this area. We need to come together as a community to set what we think are safe enough guidelines for what the minimums that perception systems might be able to provide. So that's at the car level. Now at the remote Control Center level, I think that there's these remote control centers should never be about safety in terms of the local environment of a car. So a car should always be able to reason itself out of its local safety indications. This is why we can never rely I see many of you talking about dsrc and various other communication technologies. cars can never rely on any remote communication for safe emergent actions. So for example, automated emergency braking needs to work by itself all the time. But what can remote centers do for safety? So one of the big areas in the future that we see is things like hurricane evacuation or things like there's a there's a massive pile up, there's a big accident, how to divert cars and reroute cars away from that so that we can provide faster ambulance response times. And so I think those times Have metrics will actually be able to establish at the fleet level why and when and where it pays to have fleets of various levels of autonomous cars. And and in fact, it's at what market penetration. So that's another thing that a lot of people are doing research on right now. You know, if there's just a handful of autonomous cars in the environment, that's you know that that's not going to be enough to change any major trends of safety or efficiency. So one of the things that we need to be forward looking is how many need to be out there, and what kinds of communication protocols and also what's the mix what happens when you have the Gremlins and the Pintos on the road who can't talk to anybody at all, as opposed to, you know, autonomous vehicles that do have the latest and greatest in terms of communications and cars that are anywhere in between. So for example, That might have some kind of GM OnStar capability. So I think there's a lot of research still to be done in heterogeneous, heterogeneous fleet management.
Yeah, I think from my perspective, I'm really interested in the way traditionally, it's been, unfortunately, it's been fatalities and collisions that we've we've actually assessed safety by. Michelle, I think at the forum when we had the meeting, you made a provocative point, I think you said 37 to 14,000 people die a year on the US roads. So if we want ABS to be better, what sort of reduction and we're looking for, you know, is it tenfold is 100 fold, and then we've those deaths, and how they get distributed between all the different companies. So is it really fatalities that we should be looking at when we're doing comparisons between setting that how safe is safe enough bar? Or is it something that's actually you know, adds more clarity into this safety looks at things like near miss collisions and identifies those and highlights those as leading metrics before we even Get into the lagging metrics of vitality. Michelle would be interesting. Well,
I have to just agree wholeheartedly with Missy about disengagement with the Forum has been talking a lot about that. And we do speak with regulators as well, not just yet, but also all over the world. We work with a lot of different governments on this topic. And we agree that it does need to be a partnership. And the regulators are also saying that it needs to be a partnership, and they really do want to engage with the private sector to help figure out what is the right turn to be measuring these if it isn't in disengagement, what are those metrics, and that's part of what we're doing with data for AV safety and some of our other work is to figure out what that metric should be. But when it comes to definitions of safety, as I stated before, we do need to broaden that definition that includes Not just reducing fatalities, but also looking at how we use shared space safely. So we're really advocating for a much broader definition of safety.
Yeah, I'm gonna I'm gonna I'm gonna take a quick break and have a look at the Q&A, because I haven't I haven't had look so far, because I've been concentrating. I'll see if I can pick out a few before we change topic. Oh, actually, the first one I was going to talk about next I was gonna talk about cyber security because obviously within that space at the moment, you have security operation centers, you know, these are 24 7365, running all the time checking the security of networks, constantly checking for any software update, there's made, they're continually checking. Now, I think the same will need to exist in my mind for automated vehicles, not just from the security perspective, but from a safety perspective. So when you're doing a new AV rollout of new software In the real environment, you still need to monitor that behavior, no matter how much testing you've done beforehand. It's real world behavior is what is of most concern to the general public. So I can definitely see a role from that. But yeah, just in terms of the cybersecurity, what's your views around around that, whether that's part of the AV system, or you see actually is as part of the base vehicle platform, regardless of whether it's the eiv system itself.
So I go to Michelle.
Sure. Well, um, we are we are looking at cyber resiliency. And we do look at these systems that you know, and go back to telematics systems and the ability to download software remotely into vehicles happens now and it has been happening for quite some time, whether it's wireless or once you're stopped and plugged in over Wi Fi or through some of the other main ways that you can do it. And when you do update software, obviously you need to test beforehand and once it's out that you need to continue to monitor it. So I do know that in the telematics and connected car space, there already is a culture of cyber resiliency. And that's not going to change once we move to highly automated driving systems at all. Can these systems be improved? But yeah, of course, they can all be improved and do they? Is it something that we need to keep a lookout on without a doubt. And that's also something where we see things like the auto I sack, needs to really look at going global. And also, it needs to begin looking at some of these other systems. And so that's an area of cyber resiliency that we've had on our plate of things to do at the World Economic Forum, and are planning to go forward with this summer. Probably beginning with helping municipalities look at how they can begin managing or really examining in considering Cyber resiliency when they think about some of these fleets operating in their local domains, that's probably the first place we'll begin looking at it, although we do at the forum have a platform for cybersecurity, that is, working in this space, with software as well for downloading into the vehicles.
You know, Missy, that's a it's a, it's a problem that doesn't just exist for autonomous vehicles is AI and robotics in general. You know, if you're going to start influencing decisions of these systems, which impact people's lives, potentially in a fatal manner, then cybersecurity becomes a huge risk in in all of those environments, whether that's health care or medical, and
yes, I would say, uh, one of the biggest concerning areas I have seen in my own personal experiences as an academic is the growth of adversarial machine learning. So this is still very much inside the academic space. But what what we're Doing an academia right now will eventually end up in some Russian hackers tool set very quickly. And that is the ability to trick any kind of system using machine learning, especially computer vision with a set of algorithms that basically can target vulnerabilities inside machine learning systems. And so this goes back to the inability of humans to see their own biases in the development of these algorithms. But if you haven't seen the video on YouTube, of a Tesla being tricked into going from 35 miles an hour to 85 miles an hour with a piece of tape on the 35 mile per hour sign it's really worth seeing because that is the very best illustration of what the problem is and but but it applies in other areas to the idea of face recognition. Some of the best Twitter pictures means that I've seen late lately are when makeup artists will put eyes on different parts of their faces. I know you will paint eyes in a weird place on your face that will be enough to throw off a face recognition system. And this kind of points to it points to a more broader hubris that I see in a lot of organizations, government, regular industry. It is the hubris that we want to buy the thing and if we buy the thing, and that thing can be a face recognition system, or it can be a car, but we want to believe that the buying of the thing will solve all of our problems. And so I think that we have we are living in a time where we put so much faith in technology will solve our problems, that we forget that well you know, technology, especially if you're going to start using technology that is really not mature. And this is for those of you who are following along in the QA, I've sent many links to this latest paper that I've written about the maturity or the lack thereof of AI and safety critical systems. It's just really important to remember that I am like everyone else. I want what I want now. And I'd like to have my solutions now, as we speak. I can't wait. My new iPhones are coming. They're going to be delivered today. Yay now, but unfortunately, there are some things that we should not have now. And I tell people the yardstick is if I'm still being surprised, as an academic, if people are writing papers that are blowing me away about deep learning and the fragility and the brittleness of deep learning. We may not be ready to market those technologies, whole hog across a number of industries. And so that's why I'm a real fan of the testing and evaluation because because of the autonomous nature and the stochastic nature of these technologies, normal, the regular way that we used to test technologies does not hold up anymore. So we need to find new ways to test and validate technologies, which is why I think we should all come together and a Kumbaya out in Silicon Valley and find a way to do this.
Well, that's also the heart of the World Economic Forum is exactly that is looking at what are those important regulatory or global architectures that we need to coordinate and share? So if you think about cyber resiliency, and if you think about the automotive isec, and we know that in the financial industry, there's a lot of sharing of best practices that happen, and then you look at something like I can do within the aviation industry, and you bring those together, we really do need to Be forming international global cooperative structures that allow for the private sector as well as governments to really begin sharing because it isn't going to be a one and done in this, it isn't going to be just create a product and deliver it. It's going to take creating more collaborations globally, to really manage rollout and solve a lot of these. So that's a huge part of what we're doing at the forum is really hoping to, dare I say increased globalization. Yeah, yeah. It really it really is important going forward. We're not gonna go back into being isolationist.
Yeah, I think I think is essential, which also I do a lot of work obviously through the eye to you. Especially with the UNECE forums, you know, there's a global harmonization forum so if it's for safety or for vehicle standards, so WP1 , WP29. And obviously, automated driving is a a key focus within both of those forums, they have separate work streams looking at that. And and really being able to say, Okay, look, our focus at the moment is to be on very narrow ODDs. That's where we are at the moment. But what does that look like globally? You know, will we ever get to a point where you can run a test on a vehicle in one country, and then as the convention allows you to export that product to another environment, and it be used safely within that environment? And I think there's a there's a long journey between those two things. What I also see within that forum, and I'm picking up on some regulatory comments that were made, is there's a there's a bit of a difference, actually. And there's still a bit of discussion as to whether the vehicle drives itself or software drives the vehicle. And I think maybe it'd be interesting to get your perspective on that because the human driver we can get into various different vehicles with where the same intelligent being We perceive the environment in much the same way, whether we're in one vehicle or another. But that capability for software to swap from one vehicle to another just isn't there yet. And it's a question of whether it will ever get there. That's one of the interesting points.
Yes, I was about to answer somebody and ask a related question asked if we could convert old cars to driverless cars. No, don't even try Bad, bad, bad idea. I am not a fan of aftermarket AV technologies for all kinds of reasons. But there's we're I'm not saying that that will never be a product that will sell but we are nowhere close to having any kind of aftermarket. In fact, the National Highway Traffic Safety Administration as much as as much as they drive me crazy, they actually did have a good decision. They shut one company down that was trying to do they even recognize that that's not not going to be an issue. And so, you know, I think that, now I've actually forgotten to what Where did how did we start this thing?
Yeah, I think how do you separate a driver from the vehicle?
Oh, right, right.
because I wanted to talk to you about a really interesting project that a related project that happened that I worked on with a company called Aurora flight sciences, who's now part of Boeing, we actually were able to successfully build a robotic arm that could fly one old aircraft. And then it you could pick up the robotic arm and put it in a different cockpit to fly a different aircraft. And so that's that same idea of like, Can we make a robot generalize across different vehicle types, for example, we could do that in the aviation space, because the perception task was not the critical task. So it all different about where you are in space and an aircraft is coming from so many other old mature sensors, that there didn't have to be a lot of reasoning about where the aircraft was in space, and you can't see anything out of those windows. So you know, it was fine to be able to generalize like that. But again, you know, why can't we make software that ports easily from one vehicle to another. And that's because each vehicle has its own different set of dynamics, and where the sensors have to sit where you get all the right fields of view. So within a single car, obviously, that's how companies are doing it right now. But that is actually why you can't have these add on aftermarket technologies is because what it would take to get the same world model, which is the brain of the autonomous system, what it would take to get a world model to be correct. take so much retraining. And retuning, that this just does not make for a viable product at least not right now.
Yeah. And I think that's that's the interesting thing is that the convention goes back to 1949 and 1968, depending on which country you you're resident in. And so they've lasted a very long time. And some of the behavior expectations from drivers have remained consistent. And they are the framework that enable us to move into different countries move from one country to another as humans and drive safely. So I'm, I'm really fascinated, I would say at the moment, we're in this, you know, I say generation one of autonomous vehicles were focused around the SAE language and things like automated driving system or the IDs, which encompasses both the software and the hardware. And I think that's a technical limitation of where we are missing is it's easier to take a vehicle and actually retrofit that vehicle. I know it's not the same as a calmer II approach, but it's kind of like let's take a physical vehicle and add something to the top, which is the IDs, which then has all of the sensors that were controlling the sensors and the software because there's a very tight coupling required between the two at the moment. But at that point is, what is the driver? What is responsible for making sure that it adheres to these, you know, the UN Convention on Road Traffic? Is it the software? Or is it the entire system that sits underneath it?
Well, I mean, so I hear two different questions in there. I hear a question about almost, how should we certify an autonomous driver? And which is different but related to the question of so what are humans doing in this scenario for why do we even have humans? And you know, it sounds like some weird matrix movie that needs to be made. Instead of harnessing us for our energy that our bodies put out though, the world One thing that humans where we reign supreme is the eyeball connected to the brain, right? So, you know, we are visioning systems, as fallible as we are, there is no autonomous system right now that can match our ability to see information in the world and then process it very quickly to more nuanced evolutions, like in monitoring intent. Somebody else was asking about pedestrians, you know, we can tell with very small cues, what pedestrian intent is, we do that not because not just because we see it, but because we've had so much experience. Like for example, driving around at campus where students will come out of nowhere with a skateboard, right? So I know to expect that because I have the experience and this is where that knowledge and judgment under uncertainty really comes to bear. So right now, we we, it's what we provide what the Best thing that we provide for driving is judgment under uncertainty. Just as you know, we've heard before, look, if we could get automated emergency braking, I just askance, that this is not mandated on every vehicle in every country at all times. This should just be a core technology, because we are definitely distracted and we rerun people all the time, because we're not paying attention and we know 100% that that technology will nearly eviscerate that problem. But other than that, you know, right now, humans, we still need judgment under uncertainty. And I don't care how many sensors that you mount on the car. There is I know that there's a move afoot right now to make the roof rack of sensors that you can sell and put on any different car. It's still not going to replicate judgment under uncertainty. I don't care if you have 20 LIDAR on a car. I don't care. If you Have the Best Roof Rack sensor on the car, that you are not going to solve this problem by putting more sensors on a car, because until we figure out how to at least replicate in part judgment under uncertainty, all the sensors in the world are not going to solve that problem.
You know, I like to say as well before, before I was 17 and passed my test, I, I've been a passenger in cars all my life, I've been a pedestrian, I've been a cyclist I've been on a skateboard next to the road. So all of the things in the environment that an autonomous vehicle has to deal with. I've experienced that firsthand. So when I get into the vehicle, it's I have all of that world knowledge that I'm bringing with me. And I think that can be incredibly valuable for your understanding of how the world works that surrounds you. And at the moment, we're just, you know, this a trading these vehicles but the way that we're training them isn't the same experiences that we've had on the road. So There's a very big difference between the way humans are learning has we've learned all the way through our lives, and how economists vehicles are learned.
And I think that's where the different business models and the different use cases as to where we apply highly automated driving systems is going to come into bear. I mean, the if we think about this technology is being deployed in fleets, initially low speed fleets that are delivering goods, and then we move it into shuttles that are fixed routes or in in known lanes and places, we can begin adapting and bringing these out. But the idea of dropping this technology on a lot of light passenger vehicles and then selling them in mass as we do currently with vehicles, doesn't make any sense. It doesn't make any sense from an economic perspective, or from even maintaining the machine. I do believe that in the near future, we're We're going to continue seeing these technologies deploy in a fleet operator environment, which is a fundamental shift in the way that light passenger vehicles are bought, sold operated, and the economics of it are completely different. And as we begin to learn, then we can begin rolling these out in different ways. But I do believe also with missing the importance of AI systems, and getting more and more of these advanced safety systems out there. And to that point, I think we need a lot more coordination, and quite frankly, a lot more education on our drivers as to how to use these advanced driver assistance systems that are available now. Because they can be quite confusing. They can be inappropriately named, which people have different expectations for how they operate, or don't operate. So we need to use our words very specifically, in in a unified manner around these systems, but I don't think there's any doubt that we have an operator problem when it comes to drivers. And that's us. And we need to do a lot better job training us how to drive these vehicles and communicate better.
Yeah, I think it was it was really fascinating as we've gone into the COVID-19 pandemic is really everybody's stay at home. So we've actually stopped traveling. And we've seen like a number of benefits in terms of road fatalities, but also with the environment because we've stopped traveling. So it opens up the question now of why are we traveling? Not just how are we traveling? And I think when we start to look at electric connected, autonomous shared, they've all about how we travel. They don't actually address the issue. Why are we traveling in the first place? You know?
It's a very privileged position that we have that we don't have to travel. There are 100,000 essential workers in San Francisco right now. They do have to travel and a lot of these are working Poor who do not have access to vehicle, we've done a lot of work around the Detroit area, to show that the cost of owning a private vehicle is astronomically high. And you don't have a public transportation system in place that actually is delivering something that is viable. So are already there's a huge part of the population that needs mobility. And when COVID-19 is when when some of this is behind us, and we're able to get back out and traveling. We really need to look at these broader systems of mobility, and this giant stress test that COVID has deployed on top of it, and really think about how we're moving people and what is universal basic mobility, because it's essential for economic livelihood, not just for our ability to feed ourselves and get to critical services like hospitals, but overall, we're going to have to get back to work. And that means having a functioning mobility system. And it's not one thing, it's going to be multimodal, and it needs to be better integrated. And we're going to need a lot more collaboration between the private space and the public transit operators to help figure this out. It's not going to be easy, but if we don't start having these discussions and thinking it through being really Dire Straits recovering from this economic recession that we're going through right now.
Yeah, it's it's really interesting visual, when you start to look, you know, I was looking for the UN stat. So, I read 55% of the world population live in urban areas at the moment, and that proportion is expected to increase to 68% by 2050, living in urban areas. So again, in in the of the COVID-19 pandemic Can you have Can you ever imagine a government saying, but to protect against the future spread of a pandemic, we're going to actually increase the population size, increase the density, and we're going to place you all in a city, we're going to increase the proximity of which you will live. So, at the moment, as a society, we are moving towards denser and denser city populations. But is that really where we want to be headed? You know, and how does mobility change? If we take that view that actually we want to actually increase de-urbanization, to use that expression?
Well, I'm not sure I have thought through comments on that. I can talk a little bit about rural mobility, which is an area that that we feel that a lot of, of populations have been left behind on this new mobility, innovation that we've been seeing. And by that, I mean ride hailing ride sharing on demand. And micro mobility definitely leave behind large swaths of the population and we need to be thinking about how we ensure that our rural communities have access to mobility. And it is, you know, you can argue about universal basic income, or universal basic mobility. The idea is though, you need to get people to places where they can have a high quality of life, which means employment, education, health, the basics of food, shelter, so you do need people to be able to move that includes rural populations, that includes the elderly, the disabled, and that also means looking at the different ways in which different segments of our community need to move. Obviously, women have different commute patterns than men typically do. And we need to be really thinking about this. Getting at the true data as to how people need to move and then applying the technological solutions to help deliver that. Whether that mobility on demand, whether that's adding in layers of highly automated driving systems to bring down the cost, hopefully one day, while increasing the quality of the service. Those are the types of things we need to be doing now.
Yeah. Missy, just in terms of Duke University, how is it affecting the students? What are they doing now? What's the sort of long term view on how coffee and the lockdown will affect the students?
Uh, you know, the student population I at a university, you know, in terms of the students at the University, I think they've transitioned well to the online format. But I think the thing that's really rocking their world and this has direct relevance to the autonomous mobility community is the lack of summer internships now and so So, you know, at this time in the year, we're just wrapping up classes and students would have been gearing up getting ready to go to their summer internships for various certain mobility companies around the country. And almost all of those that I'm aware of have disappeared. And so it's not that companies aren't hiring companies like Amazon is. All my students are really all my students graduating right now are going to Amazon and Seattle. So you know, people are getting hired. But it's that quasi area of summer internships where students who are still going to be students for another year or more are not able to get the summer experience that they want. And that's it's bad for the students but it's also bad for industry because then industry doesn't get to try out. That's what internships are for. Internships are a way that companies can see who the early good students are. Gonna be and who's going to get hired. And so I think there's going to be a weird, lagging year effect of a year from now we're going to see scrambling for new hires in these areas. And there's going to be a lot of movement in the space because companies will not have been able to try out people. And so I'm not, you know, I don't have the solution for how that's going to happen. How to deal with that. But I think we could see some more dramatic effect from this delayed in a year or more.
Yeah, okay. That's really interesting.
One of the things that came up I think, was, again, looking at mobility platforms in the future. And there was some, you know, innovative concepts that have been presented at CES like Toyota z palette, or Mercedes vision over netic. And these are sort of platforms where you have a skateboard and then the actual port that sits on the top can be anything from a good delivery vehicle to a passenger carrying vehicle, do you think we're going to see more of that type of future of mobility where we have some flexibility between the you know, the underlying chassis and then the thing that it carries? And that that would be a benefit in future pandemics, being able to swap out these bodies on the top for different purposes to meet the current need, you know, at the moment, the needs are very different than they were before COVID. So
I would go back, I would say to everyone, I just saw a message pop up. If you want to go into the autonomous tech space right now, what it Where would I recommend you go, I would recommend you go into delivery. There's a company out in the Bay Area, neuro I know them very well. I'm a fan of theirs. But in the spirit of conflict of interest disclosure, I have former students that are there right. So of course I know it better. But there are many other companies who are trying to come into this delivery space at CES, the consumer life tronic show I love the vehicle that could be a delivery and then you could pop put another pod on it, and then it's passenger carrying. And so I do think that those are going to be in the short term, the most successful industries that actually get products to market sooner than the broad, just general driverless car industry.
Yeah, Michelle, your views on sort of adaptive mobility platforms and how they might play a role in the future. I
love it. Are you kidding? I think it's I think it's really brilliant. And as long as we make sure that we're looking at cleaner power trains to go along with it, I think that's really essential. And that's one of the things that I worry about right here and now is the electric power trains. And what are we what are we doing going forward in the Evie space? Well, I know it's really fabulous. The gas prices are falling and some parts of the country. Obviously, it's not happening in California. Although I think I'm getting picked up on eight weeks to the gallon right now myself, we do know that we need to move forward with electric vehicles. And if we think about what can we be doing right now to help deal with this, and this will benefit highly automated vehicles as well, is Evie charging infrastructure. It's needed everywhere. It's needed in China, it's needed in Europe. It's needed across the US. I'm not convinced that we have built or really proven the true benefit for us consumers for owning an electric vehicle. But your comments, your opening comments about co2 and greenhouse gases are, we really do need to be moving forward to cleaner power trains, and we know that it's coming and we know that we need the Eevee charging infrastructure to solve it. And it is not just Put them in front of your grocery stores. Because that's not going to be the only solution. We need a really better thought out more widespread Evie infrastructure to deal with these future vehicles. hopefully there'll be fewer vehicles, because they'll be highly automated, and they can be highly utilized. But you still do need to charge them empower them. So
yeah, I think when you when you start to look at the economic downturn, that we're about to experience, Michelle wouldn't, how would that sort of suppressed the demand for your personal mobility? So owning your own vehicle? And how do you think that obviously, with oil prices as well, it's kind of it, will it? Will it start now to skew us back to it at Let's burn some fossil fuel, but actually maybe not drive so much? You know, is it going to change people's behaviors in that way?
So one of the big concerns that we have is that in this this era of physical distance and this fear for your biosafety will people Turn away from highly shared systems of mobility your your buses or subways. And so if we look at what's been happening in China right before they shut down, I think they were at about 45 days supply for new passenger vehicles. During the shutdown, they jumped up to 444 days supply. And now that the dealerships are back opening and sales are happening, sales are picking back up phenomenally. Now here in the US, we entered this with about 68 days of new vehicle inventory. I think last time I checked from Cox automotive, we're about it 98 days supply. And then if we look at the used vehicle market, there's going to be a short term glut of vehicles. So that actually the price of vehicles are going to come down and you're going to have very, very favorable financing terms for this. Which means if you're disposable in Come his cat, but you still have some confidence that you're going to be employed, you might choose to spend some of those precious dollars to buy a new vehicle, which will guarantee that you can get to where you need to go and spin safely, at least from a biosafety perspective, especially if we see public transit being services being diminished. Now, what we don't want to happen is a lot of people jumping in their cars and driving a lot more. That's not only bad for the environment, but it's also bad for safety. Go back to my numbers of 3500 people dying on our roads every day. We don't want more of that. So we need to figure out what that solution is. So we'll see it's I mean, buying a passenger vehicles is positively correlated to consumer confidence, which is positively correlated to your disposable income. So We'll see what happens in in the mid or longer term.
I think it's it's a fascinating time, Missy, what's your what's your views then on in terms of transportation in the future? How would you like to see it evolve? What do you think we need to be moving towards? Is it still electric connected, autonomous and shared? Is that still the vision that we're all aiming for the Northstar?
Well, you know, so my specialty is in the area of autonomy. I personally would love to see more electric cars. I totally agree with Michelle, that, that again, it's it's nice to think about the electric vehicle but until we have a larger, more supportive infrastructure, you know, I live in rural America, you know, it's very hard to have a V's here as a practical matter, just because of the lack of charging stations. So I would love to but it's just not gonna happen in my life because I just couldn't get around. So yes, I think electric I would move like it to be the future now whether it is that's, that's above my paygrade, the connected situation is a little trickier. I am all for better networks of connectivity. Certainly the COVID crisis has taught me just how much more bandwidth I need, and my students need I, I also have a teenager and when she's having her online classes, and I'm having my online seminar, that's a problem, right? So we definitely need better infrastructure in that way. But one of the things that I worry about in terms of the connectivity of driverless cars is that people will often have that wrong interpretation that you must have, you know, that somehow the car is doing some kind of control over some kind of connected network and and that absolutely should never be the case. And so I worry that sometimes either recently Sources get directed or people's misunderstanding causes problems and the development of the core safety technologies. Fundamentally connected networks for driverless cars are a nice to have technology, they can help with efficiency, they can help with those remote supervision issues. They can provide better feedback, they can provide for a better experience. And so I'm all for that. But before we put the, you know, the cart before the horse, I want to make sure that we get those core safety technologies addressed.
Yeah, Michelle, any last last comments before we hand back to Fred to close the session for the day? It's been fantastic, by the way, fascinating conversation so far. I've
really enjoyed it and I hear messy loud and clear on the issues of EBS. We are also I try to be powertrain neutral. I'm a big fan of hydrogen hybrid as well as electric. I just want it to be cleaner.
Yeah, okay, super. So, Fred, if you're, if you're available if you might come in back online and guiding us out of this really fascinating session. And I think everyone, thank you very much for joining the show. Missy. Thanks for your insights today. It's been a really interesting for me
Surely, and thank you so much. Thank you to Bryn, Missy, Michelle, it's been an excellent panel. Thank you for your time, learned so much. And also thank you to the participants, I saw a number of questions and active chats coming in. And just to let the participants know, this isn't the end of the show. This is part one of a series of webinars we will be doing on the concept of Smart Mobility. The AI summit normally would have taken place two weeks from now in Geneva, and Bryn, Missy and Michelle were all confirmed to come. So I'm sure you'll will see them again online somehow somewhere. I don't know exactly when yet but before the end of the year. So we're very much looking forward to that.
Uh, just to let you know, a couple of dates coming up. So two weeks from now, symbolically during that week where we would have had the summit, we will have three webinars in a row. And here we'll be tackling the breakthrough session. So if you're not familiar, but breakthrough sessions were the core of the AI summit, where we use this as a way to identify AI for Good projects and connect AI innovators with problem solvers to try and solve global challenges. So we have three topics for 2020, one is AI and gender, AI and food supply, AI and environment. And these will be taking place in the same order on Tuesday, the fifth of May, Wednesday, the sixth of May, and Friday, the eighth of May. So if you're interested in basically working, collaborating, getting in touch with AI innovators, or if you have a problem you want to solve, I would highly recommend joining these sessions because that's a step one of how you can get involved with the summit throughout the year. And with that, I would like to thank again everyone for their participation. Wish you a very nice weekend and I declare this webinar closed. Thank you