Building the Future | The outlook for automation of road transportation with Steven Shladover

Building the Future | The outlook for automation of road transportation with Steven Shladover

Show Video

- Hello, everyone. My name is Yafeng Yin, I'm a Professor and Interim Chair of the Department of Civil and Environmental Engineering at University of Michigan. So welcome to University of Michigan, building the future, distinguished lecture series. So in 2019, our department finished a two year strategic planning practice, through which we identified five strategic directions. So they're enhancing human habitat experience, shaping resource flow, adaptation, smart infrastructure finance, and automation, which is the topic of today's webinar.

This is a very exciting focus for our department, and also the profession as a whole. Our strategic visioning vision, we affirm our value proposition to our community, and reinforce our commitment as engineers working in service to society. And to highlight our vision and to formulate a work plan We launched this building the future, distinguished lecture series, which provides a forum really to discuss each direction, with aim to build a broad community that include industry, professionals, researchers, educators, and students through presentation from leading expert and panel discussion.

The series provide new insights and explore a range of perspective on each of those five strategic directions. We have a lot of support to make this vision and today's webinar come together. So we would like to thank our co sponsor of today's event. MC at University of Michigan, California PATH program at UC Berkeley and USDOT, Center for Connected, Automated Transportation. I also would like to thank the strategic implementation committee led by Professor sign up, Special thanks go to Professor Neda Masoud who is Assistant Professor here at the University of Michigan.

Professor Masoud has agreed to serve as the moderator for today's presentation and panel discussion. I also like to thank Joyce Kennedy, Michele Santillan and (indistinct), for providing technical and marketing support for today's webinar. A few more words about the accessibility of this webinar, we want to make our events accessible to all participants. So this webinar will have a live automated caption and also a transcript will be available. To choose a viewing option, click Live transcript on the control bar at the bottom of your screen, where you can show or hide the subtitles or view the full transcript.

So I will now turn the session over to Professor Neda Masoud. - Thank you, Yafeng. I would like to also welcome everybody participating in today's event.

Today's distinguished lecturer emphasizes the strategic theme of autonomy. To give an overview of this theme, I would like to play a two minute video. - Automation is becoming pervasive in our daily lives from smart homes to self driving cars. And now's the time to answer the question of what are these technologies actually gonna do for municipalities. So we believe that autonomy is the next frontier for society, we're seeing connected transportation systems that basically know where cars are, they can predict accidents and route people to safety during disasters.

We're seeing smart water systems that control themselves in response to changing conditions, we're seeing robots that are gonna build the cities of the future entirely on their own. And that doesn't just include automated construction, but includes the sourcing of the materials and the various processes in between. And we're even gonna see new sensors that are gonna be carried by construction workers to promote safety on construction sites. So we're looking forward to a future where autonomy isn't just robotics.

It's a future where the various sharing of information and data allow cities to create safer and more equitable environments for the residents. (calm music) - All right, thank you, Michele. So, during the lecture and panel discussion, I invite participants to send in your question using the Q&A function down here, and we will try to get to as many of your questions as time allows. And now I would like to introduce our various special speaker and panelist. We're very fortunate today to have a Steven Shladover, Research Engineer at UC Berkeley, who will speak on the outlook for automation of road transportation systems.

The presentation by Steve will be followed by a panel discussion with two distinguished panelists, Kara Kockelman and Reuben Sarkar. Kara is a Professor of Transportation Engineering at the University of Texas at Austin. And Reuben is the president and CEO of the American Center for Mobility.

And this brings us to today's distinguished lecture. Let me invite Steve to give his presentation. - Thank you very much. I'm pleased to be here to talk with you, at least remotely, would be nicer if we could do it in person, but hopefully, that will be possible in the future.

And I'm just gonna share my screen here so that people can see the slides. I'm gonna be talking about the outlook for the automation of road transportation systems. This is a topic I've worked on for a long time. And it's obviously a topic of considerable interest these days.

So this is a good opportunity to look at that from a longer range perspective. First, I have to figure out how to advance the slides in this mode. All right, as we get going, I'm gonna want to refer to the levels of driving automation that we have that were defined by SAE, so that we have a common language for talking about the same things. We're not talking about different concepts. And this is important to understand the difference in the roles that the human plays and the role that the automation technology plays in driving. When we have level zero, we just have collision warning systems or collision intervention systems that operate.

Yes. - [Yafeng] Maybe you want to move to the presentation mode. - Okay, yeah. Okay, thank you. So we have different levels of automation.

And people often get confused about which level of automation we are actually using at any particular time. Level one systems would allow for things like adaptive cruise control, or systems that keep the vehicle in the lane. Level two systems combine both the longitudinal control and the lateral control, but still require continuous driver supervision and driver engagement and the driving tasks. So commercially available systems, like supercruise, or the Tesla so called autopilot, all fitted in at level two category. When we get to level three, systems are able to take on the full driving task under some very limited conditions. And the driver has to be available to intervene when the system needs help, when the system cannot perform all of the tasks.

At level four, the systems will do the driving task without human intervention, and can even ensure safety when something goes wrong, even if there's no driver intervention. But they still operate only under certain conditions. Level five would be a long term dream of a system that can do everything that humans can do. So the levels zero through two are considered driving assistance systems.

The level three and above are automated driving systems. And that's what I'll concentrate on for most of the discussion. The other important concept is the operational design domain. And that is the set of conditions under which the automated driving system is capable of performing the complete dynamic driving task. And that's important to recognize because that's going to be limited for the foreseeable future. So to give the top level view of where we stand with future automation, I expect we're gonna see continued growth in the market for the driving assistance systems.

And those are gonna lead to improvements in safety as well as improvements in driving comfort and convenience. There will probably be some very limited highway applications of level three automation. Several automakers in Japan and in Germany are right on the verge of introducing some level three systems that would only operate on freeways and only in traffic jam conditions within a limited speed range.

And then maybe in the future, they'll operate over a wider speed range. There's a lot of activity on level four automation. And I'll talk quite a bit about that. But all of those are for narrow applications under limited conditions. And those are important restrictions. The impacts on transportation system are of course of great interest to everybody.

A lot of what we've seen in the media has really exaggerated what those impacts might be. But I think for the foreseeable future, those impacts are gonna be localized, localized in the locations where the systems are implemented, they're gonna lag behind the deployment, because you don't really start seeing impacts on the transportation system, until there's a significant number of vehicles that are using the automation technology. And the impacts are gonna really depend very much on how much connectivity is combined with automation. If the automation is implemented without connectivity, the impacts could be somewhat negative on traffic. If it's implemented in combination with connectivity, then there could be some much more positive impacts on traffic conditions. I do not expect to see level five automation within the foreseeable future, that's really more of a dream than a reality.

And we'll come back to that later in terms, we have to be careful to not get trapped into magical thinking, we have to be concentrating on things that are actually physically realizable within the laws of physics, and within the laws of information theory as well. This is not gonna happen, suddenly. I believe these are gonna be series of evolutionary changes that will take place over a period of years. So no revolutions. That may be disappointing to some.

But that also means there's an opportunity for young researchers today to spend their entire careers working on solving those technical challenges that still remain. So the types of level four systems, I expect we're gonna see coming in the foreseeable future begin with a lot of local package delivery, small, low speed vehicles that operate within a very narrow area, and maybe even only on sidewalks or in bicycle lanes, not sharing road space, then we should see quite a bit with long haul trucking on freeways, and the freeway environment being a lot simpler than the urban environment. I expect, we'll see some line haul transit applications and protected rights of way, things like dedicated bus ways where we could have automated vehicles. We'll see some transit feeder services in medium density areas, low speed vehicles that can get passengers to and from line haul transit systems, line haul transit stations, but limited speeds and operating in relatively limited areas.

I expect we'll see some automated ride hailing, so called robotaxis in moderate to high density, urban areas, but those are likely to be kind of spotty deployments, in limited locations, for both technical reasons and economic reasons. And I think we'll see some limited applications of freeway cruising in high end cars. So higher speed operations with automation, but in a simplified environment in freeways with relatively well regulated traffic.

So those are gonna be localized deployments of specialized applications, you're not gonna go down to your dealer and buy the car that's gonna let you drive anywhere you wanna go, that's not in the cards for long time to come. So, I realized this is a more cautious prediction than much of what you read in the media and even some of what you read in the technical literature. So let me try to explain how my experiences have shaped that perspective and led to this understanding.

So I'm gonna talk some about my technical background, both my education and my work experience, a little bit about the transportation fundamentals that I've picked up that affect this. Since I have an opportunity to work in Silicon Valley, I'd like to talk about the experience of Silicon Valley, industry ecosystem and what that means in terms of applications and transportation. And then we have to consider basic engineering principles from vehicle dynamics and control.

And I'll talk about some of my specific experience in developing road vehicle automation systems. So I spent 10 years at MIT going undergraduate through doctorate, my degrees are all in mechanical engineering with a specialization in vehicle dynamics and control. And control is a part of mechanical engineering, that's very close to what a lot of electrical engineers do as well. But I also took all of the classes that were available in the transportation systems field in civil engineering, and satisfied all the course requirements for the doctorate in civil engineering as well. So that provides a multidisciplinary perspective. And my master's and doctoral theses were on automated steering control and automated platooning of vehicles.

And that's in the 1970s. After I finished my doctorate, I spent 11 years working for Silicon Valley contract research and development company, and got deeply embedded in the Silicon Valley ecosystem. And I still live in Silicon Valley.

But then I spent 28 years at the UC Berkeley PATH program where we've done a wide range of research on intelligent transportation systems. I worked as a full time research staffer and had some program management responsibilities as well. At about four years ago, I retired from full time work there, although I continue to do some part time research at PATH, as well as some consulting and a lot of professional volunteer work. So it keeps me engaged with what's going on in the technical community.

So the work that I've done has been in a variety of different topic areas. And again, recognizing the multidisciplinary nature of this field. I've done a lot of work on developing automation systems to go on real vehicles. And they've been starting with things like automated steering, but then going on to cooperative adaptive cruise control, platooning systems and a fully automated highway system. And importantly, that work goes all the way from design through the implementation on the vehicles and the testing.

You don't really learn what works and what doesn't work until you build it, and you test it. Things that might work in simulation don't always work in the real world. I've also done a lot of work on transportation system modeling and simulation, involving formulating the models and calibrating them and validating them with real world data. And using the models to estimate the transportation system impacts of the things that we can't test at full scale, because we don't have enough vehicles available to test them at full scale. I've also done work in the transportation system planning and policy area.

And more recently in the transportation safety regulations area, where I provide technical support for the California Department of Motor Vehicles in their work on regulating the testing and the operation of automated systems on California public roads. So I mean by doing those things recognize transportation as it's really complicated socio technical system with many different elements. And it's important to understand things in all of those elements in order to be able to make progress. So on the upper left of the chart, we've got different technologies, we have vehicle technology, infrastructure technology, and information technology that are really quite different from each other, but they've got to be able to work together.

And when we're designing the systems, we have to keep in mind a wide range of considerations on the upper right, operating costs, as well as capital costs, the efficiency and the performance of the vehicles and of the transportation system as a whole. And the impact that that has in things like sustainability, and the accessibility and equity. We're up in a domain where we have many different types of organizations involved.

So we have private industry and public agencies. They have different considerations, different priorities. With private industry, the business models are critical, be done in a way that they can actually sustain a business and earn a profit.

But on the public agency side, we have to be looking at things that are meeting societal needs. Those don't always match up very well. And we're in an environment that's overseen that's really interested by the public media and the general public. They wanna know what's happening because it affects their daily lives. And those of us who are working in both the public and the private sectors have to be able to communicate with them about what we're doing so that they understand it, and they understand what the implications are for their daily travel.

We have a tension here on the supply side and the demand side of the development of these systems. And people are often tempted to just focus on one side or the other of this equilibrium, but they really have to be in balance with each other. So on the supply side, we have to consider technological feasibility, what can actually be made to work and be made to work safely in the real world, and cannot be done in a way that's economically viable, that's affordable to design and to build and to operate. And what impact is it gonna have on the transportation system in terms of traffic flow, in terms of energy consumption, in terms of emissions, and in terms of safety. Then on the demand side, we've got a different set of considerations.

We have user needs and desires and expectations, which are often different from each other. What people actually need and what they want aren't necessarily always in line. And what they expect may not be in line with what's real. We always have to consider the willingness to pay, will they really be willing to pay for some additional capability. We have to recognize people often have a distorted perception of risks.

When we're thinking of safety issues, do people perceive the risks in a realistic way. And we encounter that all the time when we have people who are afraid of flying, who don't recognize that the most dangerous part of their airport trip was driving from home to the airport, it's a lot more dangerous than once they get on the airplane. And in this field of automation, we have the added complication, that there's an awful lot of misinformation that's been floating around in the media, misinformation about the state of the art for development of automated systems, which has conditioned people to have some possibly unrealistic expectations for the technology. This is a system of systems.

And it really has to be thought about at the system level. This three part chart is something I created 30 years ago, in the early days of the intelligent transportation system field, when I noticed that people were sort of segregating into either vehicle people or infrastructure people. So vehicle, people thinking about vehicle issues and infrastructure people thinking about infrastructure issues.

And I said, it's a system, we have to be considering the vehicles and the infrastructure, as well as the people and the goods that are being moved in the vehicles, and the information that's flowing among all of them. Because the only way we can get to a well functioning system is by treating that as a system. This is a concept that's well understood in the rail, marine and air transportation fields and has been for decades. But it's been kind of a slow slog to get people in the road transportation field to think about it. That's really how intelligent transportation systems got started 30 years ago, it's trying to bring these things all together. And when we're dealing with automation, it's even more important to consider all of these together as parts of a system.

We're working in a field where the transportation becomes very dependent upon information technology. And the transportation industry and the information technology industry are very, very different from each other. And that's something I've learned by working on both sides of this. But it's important to understand some of those contrasts. In transportation, we're dealing with assets that have very long, productive lifetimes, vehicles designed to work for decades, and infrastructure designed to work for a century.

But when we're in the IT world, people think about assets with productive lifetimes of months. You expect your mobile phone is gonna be obsolete very soon, and you're gonna have a new version of software, very frequently. Transportation is really capital intensive, information technology is very intensive on skilled labor. So that creates some very different ways of doing things. In transportation, public and private sector coordination is really essential. But in the information technology, private industry tends to pretty much go on its own without much attention to government issues.

Transportation because of its public and private interaction has a very deliberative decision process. Well, in information technology, it's move fast and break things. That's the mantra. That's not necessarily good idea when you're dealing with safety-critical things in transportation. Transportation is safety-critical systems that have to be tested really carefully before they're released to the public users. Well, in IT, you're typically dealing with non safety-critical systems, and everybody expects the customers to do the beta testing.

You can't do that when it's safety-critical. And very importantly, in the transportation world, tends to be very resistant to innovations. Innovation only when pushed by external forces, whereas innovation is embedded in the DNA of the IT industry. and on the transportation side, we think about just about any major innovation that we've seen in our road vehicles, they virtually all been forced on industry by external pressures, whether it's government regulations, or foreign competition, it's been very rare for the US vehicle industry to introduce a major innovation voluntarily.

And turns out changes in the automotive market are really gradual. Typically, the fastest way of getting major change is by having a regulatory push. And I use the example here of adopting seatbelts. So from the time that seatbelt regulation was put into place, to when we actually had seatbelts in 90% of the new cars in the US took six years.

And to get it to 90% of the cars that were actually out on the road to 22 years. And then the lowest plot, the green plot here is for the actual usage of the seatbelts. So even if the vehicles are equipped with the technology, that doesn't necessarily mean that people are gonna use it. So we're in a system that's got some very long timescales and in their natural evolution. Similarly, in the growth of popular automotive options, features that we now take for granted on vehicles got into the vehicle fleet for gradually. So this figures on a 35 year timescale, and we look at the percentage of new vehicles sold each year that are equipped in the top plot automatic transmission, but then power steering, air conditioning, disc brakes, radial tires, electronic ignition, these things take decades to go from being options on high end vehicles to being standard equipment on all vehicles.

And then we still have to consider the turnover of the fleet after that. It's also important, the ones that are slower, or the ones that are sort of more expensive, and then involves a bigger change to the vehicles. So when we think about automation, that's a really, really expensive thing. And it makes a really big change in the way the vehicles operate.

So it's likely to be on the slower end of this type of a market growth curve rather than the faster end. So thinking back on the kind of engineering experiences I've had, there are quite a few lessons to learn from that. And I've summarized a few of them here. First, Murphy's Law is inescapable, things will go wrong. And you always have to assume that the worst kind of failure will occur. It may not occur tomorrow, but it'll occur the next day.

So when you're designing the system, you always have to think about what are those worst case scenarios that your system has to be able to handle. And if it can't, you're in trouble. Also, full scale vehicle experiments are really expensive, they're really time consuming, but they're essential. And I said they're worth vastly more than simulations.

Think of simulations of being a little bit more like the fiction part of the bookstore, while the full scale vehicle experiments are the non fiction part, you can't fake it in a full scale vehicle experiment. So it has to really work. Well in simulation, you can make anything work if you put the right assumptions into it. And I've done lots and lots of simulation work, but simulations are still crude approximations to reality. And it's very easy for people to kind of ignore that and think that the simulation is the reality. It's not, it's a set of assumptions that have been put together, and they've been put into computer code, but they're not the reality.

And if the simulations have not been validated, the results are kind of worthless. I know that's a strong statement, but it just reflects the assumptions that you put into it, rather than reflecting a representation of reality. And if you're trying to predict what's gonna happen in the real world, the simulations really do need to be a representation of reality. When we're dealing with safety-critical systems, all of these effects get amplified dramatically.

So you may be able to get away with some of these things when it's not a safety-critical system. But when a safety-critical system, this becomes really important. So after working on this road vehicle automation field for 50 years, there are quite a few lessons that I've learned from that that I'd like to share here. And I recognized a few years ago that companies working in this field need to spend at least a decade of effort and some billions of dollars to get to the point of understanding, how much they still don't know. It's only after they've been through that decade plus and those billions of dollars, that they start to appreciate just how complicated and difficult this is. And they start making more sober predictions about what's really gonna happen in the future.

So you see that the most aggressive predictions about what's gonna happen typically come from the least experienced companies. And you have to be very careful about the media reports, because most of them just recycle the press releases that companies put out that are full of exaggerations and the represents their hopes for the future, rather than the realities. There aren't very many of the media outlets or media reporters who look at these things critically, and try to separate what's real from what's not. The complexity of driving, automated driving greatly exceeds the complexity of aircraft autopilot automation.

So people talking about oh, we can do it for airplanes, why can't we do it for cars? Well, it's because the road driving environment is vastly more complicated than the aviation environment. The other thing that's important to understand is, human driving safety is already extremely high. And when we look at real numbers, and if we're talking about an automation system that's got to be safer than a human driver, that's a very, very high bar to reach. And I'll show some numbers in a little while to explain the reasoning behind that. When we're assessing the progress in automation systems, we really need to focus on the failure rates rather than the success rates. And again, I'll explain that a little bit more.

Because the failure rates give you a much clearer sense of where you are. And finally, in all of this safety assurance, I would describe as the central challenge. How you can design the system, and then test and demonstrate the system to the point that you can ensure that it really is going to be safe. That's the primary impediment to getting systems deployed. So we have a really extensive history of false claims that we have to work out from.

And I just pulled some out of my files, going back to about 2013, 2014, to get some of the news headlines. So the predictions that people were making about what was gonna happen by the end of the 2010's decade. Well, now that we're a couple of years past the end of that decade, we know that none of that stuff actually happened. But unfortunately, that was the prevailing wisdom through I would say about 2018, the hype peak was probably around 2015 or '16, when everybody thought by 2020, you know, cars with human drivers were gonna be obsolete and everybody was gonna be automated, and obviously, it didn't happen. Now, we know that that was not true. So we have to be very careful in assessing those media claims.

Just a little bit about the comparison with aircraft auto pilot automation, if you're flying an airplane up at 30,000 feet, you don't have to keep track of very many other objects that you might crash with. You've got loads of time to respond if anything goes wrong. If you're out on the road, you need to know the locations and the speeds of many other entities, other vehicles, pedestrians, cyclists, everything else in the area, that might be a hazard to a very high accuracy. And if you don't know what to do within about a 10th of a second, you're gonna be in trouble. If you're flying at 30,000 feet, you've probably got tens of seconds to be able to respond.

So the technical challenges of avoiding crashes with other entities are far more complicated when we're in the road environment than in the aviation environment. We want to talk about this safety baseline too. I often hear people say, oh, humans are terrible drivers, we have so many crashes.

Well, not really when you look at the exposure. And when you look at the rate of occurrence of those crashes. If you just take the current US traffic safety statistics, you see, we've got three and a half million vehicle hours between fatal crashes. And if you took that three and a half million hours represents 390 years of continuous 24/7 driving.

And if you look at injury crashes, it would still be equivalent to seven years of continuous 24/7 driving. That's only gonna get better as we have more use of collision warning and collision avoidance systems, so human drivers will be getting safer and safer. But if you think about these 390 years of these seven years of continuous crash free operation, how do you compare that with your laptop or your tablet or your smartphone? Try to imagine that any of those devices being able to run for many years without a serious glitch. So that gives a first indication of just how difficult this is. And then unfortunately, you can't test your way out of this, you can't test your way to prove it.

RAND did study a number of years ago showing that the amount of testing that you would have to do to be able to prove that the system was at least this safe, is many times larger than these numbers. So the technology for automation is still relatively immature compared to that kind of a safety target. The perception technologies remain below the perception capabilities as skilled human drivers. We don't have viable methods for software safety and security protection.

So you can't design software that has such a low level of bugs that it's gonna be safe. And when we look at the results of real world testing, we see that systems are still far behind those human driver performance. Use the example of the disengagement reports from California. That's really the only data that we have at this point about the performance of highly automated systems in real world traffic.

Now, there are a lot of problems with disengagement reports, they have to be treated very carefully and interpreted with a lot of care. But what we have on the plot here are the disengagement intervals for the best of the companies that have been doing the testing in California since 2015. And this is on a large scale.

So you see this is the number of miles in between events. And in these cases, the events are disengagements, that had to be done in order to avoid an adverse safety event. Now the companies don't report the specifics of exactly what kind of event they were avoiding. But the companies whose data are shown here are all using simulations, counterfactual simulations to play through the scenarios.

And they predict that these disengagements would have led to a bad outcome. So all of the unnecessary disengagements were taken out. And the lines up above show the rate of occurrence of the minor property damage crashes, the reportable property damage crashes, injury crashes, and fatal crashes. And now you can see, the companies with the most robust systems are still some distance away from even getting up to a disengagement rate that matches with the minor property damage crashes.

Also make a note of one line that you see that has a negative slope between 2020 and 2021. And that's very instructive one, because that's an example of a company that during that year, switched from a moderately complicated operating environment to a very complicated operating environment. And they switched from one vehicle platform to another vehicle platform. And this illustrates the scalability challenge. So you get a system working really well, on one particular vehicle platform in one particular environment, you go to a new environment and a new platform, you've got to relearn a lot, there's a whole lot of work that has to be done to get your disengagement rate back up to a higher level. And by the way, many of the companies that are testing are way down near the bottom of this plot, didn't even try plotting the bad ones, I just plotted the best of them.

And if you consider something like the Tesla Autopilot, that would be way down at the bottom of this as well. So this is really a little bit like climbing Mount Everest, and you have to be careful that you don't look at how close you got to the top of Mount Everest if you have to, because that's like I'm gonna go and climb Mount Everest. And when I get off the airplane in New Delhi, I've covered 90% of the distance from San Francisco to Mount Everest Summit. And I fly to Kathmandu and I've covered 99% of the distance to the top of Mount Everest, but it doesn't really matter, anybody can do that. All the really hard work comes at the end.

Because look at the left side of this chart, if you're going to be equivalent to a human driver, you've got to handle 99 and about seven or eight nines after the decimal place of the scenarios that you encounter, just to match what human drivers are doing. So that's why we need to focus on the failure rates because the failure rates are the things that go into the crash statistics. Doesn't matter if you're 99, or 99.9% of the way there, what matters is what's your failure rate? Is it one in 100,000 miles or one in a million miles. And that's a factor of 10.

And turns out, it's at least a factor of 10 in effort. Because as you get to those higher and higher numbers, you're having to deal with more rare and more complicated hazard scenarios. You've already handled all the easy ones, now you have to deal with the hard ones. And it gets harder and harder, just like the mountain gets steeper and steeper as you get closer to the top of the mountain. So this is where it safety assurance becomes the central challenge. This is where you bring together how you make the technology safe, how your regulations have to govern the safety requirements, and how you get the public acceptance as well, because you have to be able to communicate the safety effectively to the public, in order for people to be comfortable using the system or sharing their roads with the system.

So we've got several challenges in getting to automated driving safety. Is first the technical challenge of how can it be designed and developed to achieve safer driving than humans? What has to go, what kind of technology both sensors and hardware and software? And then assuming gone through that design and development process, what's the process that you go through to produce a credible assurance of that safety, without unduly jeopardizing the intellectual property of the companies that did all that work that put it all together? So how do you produce a credible assurance revealing enough to the public to earn their confidence without revealing trade secrets? And what kind of outputs have to be presented to the public to earn that public trust to facilitate the acceptance, and what kind of a regulatory approach is then been needed to facilitate the trust in the ones that are good, and to constrain the behavior of the bad actors in the industry, and we need to be conscious, there are bad actors in the industry. I've had enough opportunity to see what some of the companies are doing to know that some companies are doing a really good job with safety and some companies are really doing a not good job in dealing with the safety issues.

So to get to safety assurance process that's widely accepted is gonna be necessary to agree on what the reference baseline is, what are you comparing against? What's the requirement that has to be met? How safe is safe enough compared to the baseline? Is that average human driving? Or is it highly skilled human driving? Or is it a baseline based on rail or air transportation? What methods need to be applied to compare with the safety baseline? And there are a lot of people working on scenario-based assessments, those have some big, big technical problems that are gonna have to be solved in terms of selection of scenarios, identification, what are the relevance scenarios, and then how much do you test and how much do you simulate? There are gonna be challenges about the relative roles of the developers of the automated driving systems, roles of independent evaluators and roles of government regulators in coming together to get to the point of being able say, yes, this one's safe that other ones not safe enough. And how do we extrapolate from some small scale scenario-based safety assessments to compare with the current statistics that we have on real world crash safety. So, looking ahead, there are lots of opportunities and challenges here in automated road transportation, but it's gonna be a long, winding road.

This technology is still in its infancy, even though there's been billions and billions of dollars invested already. So don't expect to see a lot of highly automated vehicles in the near future. But the plus side of that is, all of you who are early in your careers will be able to have a full career of interesting work to do on this to try to solve those problems that still remain.

And with that, I thank you for your attention and welcome getting into the discussion. - Thank you, Steve for the outstanding and really thought provoking presentation. It is a great time now to transition to our panel discussion and hear from Kara and Reuben. Great. So let us start the panel discussion by just hearing a little bit from Kara and Reuben on your take on Steve's presentation. Kara, would you like to get started? - Sure.

Yeah, there's lots of excellent points being made by Steve. Like the idea of the supply side, a lot of people worry about demand side, I'm not really worried about demand side. But, you know, once some of these manufacturers get a really strong vehicle design ready, just gearing up the assembly lines to keep up with demand, and so that's gonna keep prices high. And I assume the fleet managers, so the shared autonomous vehicle fleet managers will probably gobble up a lot of that, although they are pretty savvy in how they bid for these things. But I think they just, you won't be able to roll them out fast enough, once you do have some solid designs. And we see that you know, with Tesla's EVs, you know, the prices jumping up, because they just can't produce them fast enough.

And so I'm really glad you mentioned the supply side. And yeah, I think if you know, adoption is not a big issue if the prices are less than, let's say, 10 to $20,000 per vehicle. And those will fall as they but I just don't think they can supply as fast as demand is going to request them. And a lot of that adoption comes from just seeing others using these technologies safely.

But you know, sometimes a lot of that comes from maybe low speed or low complexity environments. I know that, you know, our vehicles keep us incredibly safe inside the vehicle. But it's the people on the outside of the vehicle that are so vulnerable, at almost any speed, especially if you hit somebody who's over age 65.

And they just, they really do not have a good chance of surviving so many of those crashes, and they're very hard to spot. A lot of the automated braking is not working very well for them at this point. And so I'd really love to see much more automation on our vehicles right now to protect vulnerable users. We're pretty well protected inside unless we do crazy things. And I don't know why we're not mandating speed governors. I don't think any vehicle should go more than 80 miles per hour.

I set my Tesla to do that the other day, and then my husband changed it. Crazy, he's thinking like, he's gonna be in West Texas sometime, and he's gonna want to go 85 I mean, that's not gonna happen for months, it's so easy to remove that governor, and yet he did not want it on there. So we're kind of our own worst enemies, which Steve was talking about, we don't really appreciate risk properly.

You know, sort of like with the COVID crisis, and so many people not masking, I think we're seeing more people now not using their seatbelts. And Steve had that excellent image of seatbelt adoption, always lagging, you know, seatbelt availability. And, yeah, also like this Mount Everest comparison, that's so neat, especially when you're talking about, you know, an order of magnitude change in the crash rate. But I do think we will achieve it.

I think the cost of if human driving is just so high, even if they didn't reduce the crash rate at all that $1 trillion, a year of savings by maybe reducing the burden on the driver, by maybe 50%. So if your value of time, you know, is on the order of 15 or $20 an hour, just reducing that so that you can sleep, which is something I would like to do in round instead of, you know, try not to hit other people for 20 minutes, or whatever the duration of your trip is out. I think that's incredibly valuable. So there's obviously great motivation for this and that's why so many manufacturers are spending a lot of money on this. But yeah, I'd love to see a lot more automation than we already have, doesn't have to be self driving. It just has to be basic stuff.

It's sort of a moral or immoral I should say, to not have automated braking required. And then I do think that we will probably avoid tens of thousands of deaths but part of that's just by keeping speeds down, we could do that today, if we just put you know, governors, we could achieve a lot of fatality and injury savings, and probably lots of property damage crash type savings right away. And I do think you know, there will be a lot of value to having a vehicle fleet that is managed by someone else to save us some of the hassles if they come quickly, which our shared autonomous vehicle fleet simulation suggests that they will even in the rural locations, which goes at say one of John Niles's questions that are in the Q&A box there. So I do wanna remind people that a vehicle is a loaded weapon and even though they don't die as much as you might expect out there, the vulnerable users are I think now 20% of all lives taken in this country.

And so we really do need to make these vehicles much safer for those around them, not just inside. - Thank you so much, you hit a lot of important points, thank you. Reuben, I would love to hear your take. - Thank you. I would also say that in Steve's presentation, there's probably a takeaway on each slide that I have had, in one of my presentations. I mean, there's something to walk away with on each of those slides.

I'm just gonna hit some of the high points on some of the things that resonated with me. One, that this is gonna be an evolutionary versus revolutionary change. And that we will see pockets of revolutionary things happen, things that demonstrate the art of the possible excite us, but those are gonna be in limited and controlled situations. And really, it's gonna be a slow and steady progression of more and more technologies, getting into vehicles and automated emergency braking is a big one, forward collision warnings, increasing level two automation, and so forth. I think by the time we get to a level four, or even level five personal ownership of a car, it's gonna be kind of anti climatic, because we're gonna have been exposed to these technologies for a long period of time.

And so by the time we get it, it may not be super exciting, but it will be transformative, and we look backwards on those decades and see how much has changed. So forward looking, it's gonna seem like it's taking a long time, in reverse, is gonna look like the whole landscape has changed. I do like also the analogy of the 35 year penetration rates into turning over a fleet.

That's been our experience in Berlin in easy technology for putting fuel injectors in cars took 16 years to transition from carburetted engines and another 17 years to turn over the fleet. There are some people that say that automation with robotaxis will eventually diminish the need for 240 million vehicle fleet and that the mileage will accelerate how fast it turns over. I think that'll take the tail end of the transition anyways, it's still gonna take a number of decades to get there.

But look at how far we've come versus look at how far we need to go as a typical thing in any sort of transformational entrepreneurial area. Because it is challenging to look at, you know, how far you come from zero to where you are. But again, I think, you know, there's still a long way to go.

And I think that people are realizing that the RAND study is something that's often referenced. And I would say just one caveat to the RAND study, because it does say I think 5 billion miles of driving, which is the distance to Neptune and back to prove that an automated vehicle is 20% more safe than a human driver. That's to prove it, it doesn't mean that that's what it takes to develop it. So you can release technologies earlier than all those miles. And they could be safer, you just have to be careful about how that deployment takes place, because you haven't proven it yet. And that ties into then that last bit, the last last piece, which is failure is gonna occur and the worst kinds of failures are going to occur.

And it shouldn't put you at bay to not follow these life saving technologies or potentially life saving technologies. But you have to be responsible then with your deployment proactive. And I think by not doing so you slow the process, because it has a chilling effect with, you know, Congress in terms of passing laws to enable deployment of these technologies. It has a chilling effect with agencies and with consumers. And so it's really the companies that really think about proactively, are looking at it not only from a due diligence perspective on the front end, but when something happens, how do they analyze and understand that? How do they communicate that back to, you know, legislator stakeholders. I think that that's gonna be very important, because again, to not move forward on these technologies is to have 40,000 deaths or an increasing number of deaths happening, right.

And as Kara mentioned, maybe there's other technologies that can mitigate that with speed, and so forth. So a lot of what Steve said, I think, is really spot on. And I think just some of the things I highlighted were the ones that resonated with me, (indistinct). - Hey, thank you so much, Reuben. We do have a number of questions that were submitted by the audience at the time of registration, and I would love to hear your thoughts on them.

The first question is, how well do we need a vehicle to cooperate with human drivers to keep safe? - I think it's gonna be essential if they can't cooperate with human drivers, they're not going to be part of an effective traffic system, and they're gonna create more traffic conflict. And that's a big challenge that the developers face when they need to understand the driving protocols that are particular to specific locations. So they might develop a system that works well in one particular location. They take it to a different location and find the local driving driver behaviors are different. And now they have to change their software to conform with the different driver behaviors in another location. So that's one of the big challenges everybody has to deal with when they try to expand beyond their initial deployment side.

- Yeah, thank you. Reuben, please (indistinct). - Back into that same question in a little differently way. How the EVs interact with other non EVs. And then there's how these technologies interact with the driver sitting in the seat.

And with level three coming out, driver monitoring systems are gonna be very important in terms of making sure there's a proper level of engagement. So it has to interact with the driver inside the vehicle as well as it does with the vehicle to vehicle interactions. - Kara, would you like to add to the conversation? - So I think the question is about autonomous vehicles interacting with human driven vehicles. So it's not a quasi autonomous vehicle interacting with its own driver.

So based on that presumption, yeah, you don't want them acting too differently, (indistinct) taking advantage of the other, which you know, can happen, there can be impatience, and that's has led to a lot of these collisions, I think, for (indistinct) the following driver just disagree with the slow speed choices. So yeah, they can definitely cause a lot more problems than they solve if they drive erratically or slowly. And of course the manufacturers don't want them crashing. So they're doing everything they can, including slowing down to avoid that and stay out of the headlines.

- Hey, thank you. Thanks to all of you. I think we can cover one more question. What policy shifts are needed for an automated future? - I'm actually gonna say, I don't think there are need for policy shifts. I think they need to be implementable within the policy framework that we have already.

And it's more a matter of what needs to be done to demonstrate that they're really safe enough that they merit sharing that road space with everybody else, and that they're consistent enough in their behavior with everybody else. Here in my neighborhood, I coexist with automated prototype vehicles all the time, and you learn, if you come up behind one of the stop sign, it's gonna sit there at the stop sign for three seconds, even if there's no traffic visible on any of the other approaches before it starts moving again. And if you're gonna be impatient and expect it to do a rolling stop, you're gonna wind up hitting the back of that car. But they have to be able to fit in with the rest of the transportation system.

- Thank you. - If I could chime in there. So I do think that there are people looking at the policy landscape at a federal state local than an interstate level. Alliance for automotive innovation put out a 14 point policy roadmap, but some of the guidances, they wanna maintain the separation between state and federal roles, maintaining design, construction, engineering, FMVSS requirements at the federal level, and then maintaining operation and licensing regulations at the state level. There's a lot of states that are already implemented, or even reduce restrictions to allow autonomous vehicles to be on the roads, managing ordinances across local jurisdictions, so that you don't have to program a vehicle to operate differently as it goes city to city. And then interstate harmonization of regulations and infrastructure are also gonna be important.

What I've heard a lot of and not everybody agrees with is that people want as few restrictions as possible for putting out these technologies, but that they want to have the right level of legislation to enable safe operation. And so at the federal level, they do need to change for novel designs of vehicles. If FMVSS requires steering wheel be in the car that there'll be mirrors and a driver on the seat and that next generation vehicle is gonna be different. There is a specific request for them to at least create a separate set of rules around design of novel vehicles. Otherwise, you can't self certify against a standard that assumes the driver is gonna be in the driver's seat. - Thank you.

- Yeah, there are some standards like that that might need to be changed. But I believe the Alliance proposal would be a catastrophe in terms of safety, because the Alliance proposal would basically turn it into the wild west, and would leave the behavior of the vehicles essentially unregulated. So that's gonna be a source of a lot of contention along the way. - Kara, would you like to have the last word on this question? - No, I don't think there are big policy needs. I mean, I do think California was smart to require this engagement and details and stuff like that.

So I'm not sure you know, why others don't have that. But yeah, I think we do need data and I would love to see that willingly given over. So that's really the only policy issue and of course, freeing up some of the the archaic licensing language that you know, we've talked about pedals or steering wheels and you know, human driver and a seat, those kinds of things are pretty easily changed, I think. And many states are leading and that's great to have some demonstrating for the rest of the world.

I think China's gonna be a big demonstration for us and that's gonna give us a lot more confidence as we go forward. - Hey, nice, thank you. Unfortunately we are approaching 4:00 pm.

So we have to conclude today's event. Let me thank everyone for attending and also give a special thank you to Steve, Kara, and Reuben for the engaging discussion today. Now, a few remarks before we close today's session, please visit umservicetosociety.org to explore our strategic directions. And also please let us know what you thought about today's event by responding to a short survey which we are adding to a link here. Here it is right now.

Please remember to register for our upcoming webinar on smart infrastructure finance on March 18, where Peter Adrians will be our featured speaker. Thank you again everyone for attending, and I look forward to seeing you in the next one.

2022-03-01 12:34

Show Video

Other news