The Road to Vehicle Automation: How Far Are We?

- Welcome to another episode of "Supply Chain Frontiers," the MIT CTL podcast where we explore the trends, technologies and innovations shaping the future of supply chain management. I'm your host for this episode, Mackenzie Berry. Today, we're diving into the world of Advanced Vehicle Technology and its implications for supply chains and mobility. Joining me are leading experts from the MIT Center for Transportation & Logistics, Dr. Bryan Reimer, Founder and Co-director of the Advanced Vehicle Technology Consortium, Dr. Pnina Gershon, Co-director of the AVT Consortium, and Bruce Mehler, Co-director of the AVT Consortium, which is commemorating its 10th anniversary this year. Here's what I know. AVT was founded with the aim of developing new data that contributes to automotive manufacturers', suppliers' and insurers' real-world understanding of how drivers use and respond to increasingly sophisticated vehicle technologies such as assistive and automated driving while accelerating the applied insight needed to advance design and development. The Consortium has developed a deep understanding of driver behavior and consumer preferences with technologies like Tesla's Autopilot, GM's Super Cruise and Ford's BlueCruise, as well as many other driver assistance and support features. Insights from AVT aim to help organizations better design and market products more closely aligning with real-world consumer use while advancing safe, convenient and sustainable mobility. In this episode, Bryan, Pnina and Bruce will explore key insights from their recent research, the role of data in shaping safer and smarter mobility solutions, and how the Consortium is addressing critical questions around driver behavior, automation readiness, and industry collaboration as they celebrate their anniversary milestone. So welcome to the episode. I'm so excited to have you all. I introduced your titles, but I would love if we could go around and you all introduce yourself with the work that you'll do in the AVT Consortium.

- Once again, my name is Bruce Mehler, I'm a Research Scientist at the Center for Transportation & Logistics. My own background is in psychology and physiology, and I've worked actually in the medical device industry for about 22 years before coming to MIT. So I do have some appreciation of the practical side of industry in terms of actually having to develop and produce product. And I think that's a really important aspect of working in a group like this at MIT, where we're interested in both high-quality academic work that provides theoretical insights while at the same time investing energy and actually translating those insights into operational information that industry can actually use. And so I came to MIT about close to 20 years ago now and have been working in this area since that time.

- Mackenzie, thanks for having us. Bryan Reamer, I've been a Research Scientist within the Center for Transportation & Logistics for about 22 years now. I'm known as a pragmatic possibilist, studying the intersection of behavior, technology and policy at really the intersection of topics around automation, assistive driving and AI. Really focused on understanding how we can encourage business innovation from research, moving concept from the lab to product innovations that touch us. How do we create the ecosystems that allow us to engage in the technologies that can excite and delight the customer and impact how we live and move and leverage advanced technologies as part of our daily lives? So a lot of my background comes from the study of driver behavior and the evolution of technology in the auto sector over the last two and a half decades.

- Hi, so my name is Pnina Gershon, and I'm a Research Scientist at the MIT AgeLab, where I study how humans and technology interact and how these interactions impact mobility and driving safety. I Co-direct the Advanced Vehicle Technology Consortium, leading the research on naturalistic driving. So I bring many years of experience in data science, AI, driving simulation, computational modeling, and human physiology. Before joining the AgeLab at MIT, I was working on driving safety for high-risk population at the National Institutes of Health, and before that I was focused on tactile and haptic interactions.

- Wonderful. Bryan, I introduced the Consortium a bit, but better to hear in your own words. Can you give us a brief overview of the Consortium, how it started and what its primary goals are?

- Sure, Mackenzie. What we discovered with our industry partners was that there was a lack of information that existed in federal and research databases around how consumers were touching data in real cars once they took ownership of the dealer. So cars designed, developed with incredible amounts of research, investment, product development, product planning, packaging, and ultimately moving a car to the dealer. But once the keys are handed to the consumer, there's very little understood, and that white space was what AVT was built upon. It was built as an academic industry collaboration informed by industry for industry. How do we derive the insight in what people are leveraging in their own lives to help enhance the development of product, the impact of the products that are being developed, and the policies that will long-term govern this industry forward? It was captivated as Advanced Vehicle Technology because at that time we knew automated, assisted were narrowly focused. We wanna be thinking about the wide swath of technology that over the course of several decades may be shaping how we live and move, so very much the mobility story much more broadly.

- And if you could take us behind the scenes a little bit, what methodologies and research techniques does the Consortium use to study driver behavior?

- So at the AVT Consortium, we focus heavily on naturalistic driving studies. That means we are collecting data from real people driving real roads in their own environments and doing their regular routine. So we take two main approaches for this. The first we look at long-term study of volunteers, people that allow us to instrument their own vehicles with data acquisition systems. Many of these participants have been with us for years now, and that gives us a rare opportunity to look at how behavior evolves over time, so months and years. We can study how people adapt to new features, but also how they respond to software updates and how trust in automation evolves over these long periods of time. Second, we run field operational tests, which basically we have our own vehicle fleet, the MIT AgeLab vehicles, in this case we loan our vehicles like we have Volvo S90s, Tesla Model 3s, Cadillac CT6, and so on. So all of them are equipped with Advanced Vehicle Technologies, and participants use these cars just like as if they would use their own vehicles for their daily routine. Now that's a great way to learn how novices or naive users get used to automation or interact with automation. So we look closely at early learning and first impressions and how people explore or avoid using advanced technologies.

- Yeah, interesting. Not only how they use it, but what they avoid using as well, which can tell you a lot. Now, I imagine, and I'm sure our listeners as well, that you're dealing with a lot of data here. So curious, what are some of the biggest challenges in working with large-scale naturalistic driving data and how do you ensure data quality and reliability?

- Well, I think that, as you said in the question, the biggest challenge around working with naturalistic data is that it can be very, very large. Particularly in our case, depending on the vehicle that we're working with, we're using anywhere from three to four cameras in the vehicle all the time. And so if you can imagine that literally you have high resolution cameras, three or four in each vehicle collecting data every time the person goes and drives for a month or two months or three months or longer in time, that data really builds up quickly. And actually one of the things that allows us to pragmatically do the work that we're doing is because of our investment through MIT and through the state in the Massachusetts High Performance Computing Center, we actually have access to a whole farm of servers and large data storage. In fact, if we didn't have access to that, we could not work with the data that's as large as we're using. In fact, when we first started doing this work and before we moved to the High Performance Computing Center, we were basically the second largest user of storage space here on the MIT Campus. I think astrophysics was the largest, and all the rest of us were small by size, but we were number two for a long time. So once again, the biggest challenge really is that size of the data, processing the data, storing the data, curating the data. And I think in terms of data quality and reliability insurance, my colleagues Bryan or Pnina may have something to add on that.

- Yeah, I mean that's a great question. Honestly, this is something that we think about every day when you're collecting this huge amount of data and mixed data, including video, audio, vehicle signals, GPS, everything that Bruce just mentioned. You're not just managing the technical complexity, but you also have the responsibility to protect the people that volunteered to share their driving with you. So a lot of our effort is really technical, but also basically making sure that we have quality assurance processes in place that combine statistical tools and human efforts to ensure that the data we use for research is intact and safe.

- I think one thing to add is that the size of the datasets that we collect are in the hundreds of terabytes range, which by industry standards is modest, by academic standards is extremely large. It requires software engineering, data security, computational investments that are not traditionally found on college campuses. And as Bruce mentioned earlier, it's really MIT's forward thinking investments with the state of Massachusetts, a few of the old colleges, universities that have really enabled AVT to be what it is.

- And speaking of that, you all just celebrated your 10th anniversary, which is a major milestone, celebrating it this year, and held an event here on campus which brought together major stakeholders from across the automotive industry. Curious, and if you could share your biggest takeaway from the event.

- I think it was a sentinel event. I think the goal that we had was to really move the conversation out of the analytical elements that AVT has been focusing on or a decade and move it back up a level or two to the strategic level. And the panels and keynotes are all available off of our website, so in long form. But it really shared with me and highlighted that the white space in automotive safety research is shifting. There's been a lot of focus on the use of technology to try to encourage and improve driver behavior, but there is a lot of new elements that technology is uncovering here. And the repair vehicles getting to be too expensive per se, we're totaling way too many vehicles. The context of sensor calibration, when you do have a car that needs this windshield replaced, the amount of hours of labor that needs to go into calibrating those sensors so they actually work correctly again and the unknowns of a used vehicle that you may purchase and whether the sensors were actually calibrated correctly or the consumer found a runaround to a cheaper repair facility. Conversation with John Bozzella, CEO of Auto Innovators, and Mark Rosekind, former Administrator of NITSA, really focusing on the complexities of vehicle safety policy looking forward. But I think the most important takeaway and discussion point is that the United States and the Western world no longer leads in the area of Advanced Vehicle Technologies, China does. The explosion of the Chinese auto industry in the last five or more years and the movement and acceleration of assisted and automated driving features into the fleet is astonishing. The Chinese government back in late April, beginning to crack down and set some standards around assisted driving that really lead the world, thinking about the innovation in China being curtailed by regulation to a degree that in the United States and most of the European nations hasn't been touched on yet. So when we look forward, we need to be thinking from our perspective here at MIT and in the United States of how do we accelerate forward? And I believe very strongly that collaborations and partnerships are where we're gonna have to get there. We're gonna have to bring industry and academia, industry and industry back together to recapture our leadership in this area quickly.

- Well, and as I understand it, Bruce, you attended the Shanghai Auto Show as well as the Detroit one. What did you notice in terms of difference between the two that Bryan had just spoke to?

- It was totally eye-opening. As you said, I went to the Detroit Auto Show in January, and it was a nice experience, lots of interesting new cars on display, lots of interesting booths and so on. And then about two weeks before our event in May, I was in Shanghai and went to the Auto Show there and was bowled over. The Auto Show in terms of space was at least three to four times larger than the Detroit Auto Show. The number of vehicles were three to four times those that were present in the Detroit Auto Show. And it was amazing looking at the evolution that's taken place, not only in the vehicles themselves, but the relationship between the Chinese automakers and the European automakers. So in fact, there were a number of companies there presenting who did have partnerships with the European automakers. And you could see what has changed over time from the European automakers being the leaders and having some slightly less expensive versions produced in China. It's really turned around where the Chinese automakers who have still partnerships were actually the predominant people at the show, and their vehicles were larger, fancier, equipped with more features than their European partners. So clearly a very rapidly changing space, very much as Bryan has indicated. One of the things that did come out from both being at the Shanghai Auto Show, but I also spent a couple weeks at Tongji University in Shanghai, which is really the leading university for automotive studies in China. And one of the interesting things that came out from that is talking to colleagues there about the work that they were doing. One of the areas that did stand out where we are still leading in a sense is that naturalistic data collection in real vehicles by customers on the road is relatively unknown at this point. And so in fact, the AVT efforts in that particular area of studying human interaction on the real road with real drivers is still an area where we are leading, and it's important to continue that emphasis.

- Yeah, absolutely. And turning to public perspective, I imagine you all think a lot about this in your work and maybe have to dispel some things among people that you know right in conversation with the expertise that you have, but what are some misconceptions about vehicle automation and driver behavior that you encounter most often?

- So one of the biggest misconception is that automation makes driving passive. In reality, most systems still require an active driver and active supervision, and drivers often do interact with the automation. Another is a belief that more automation necessarily mean more safety. So driver behavior, trust, and how systems are actually used play a huge role in the outcome. And automation can support the driver, but it doesn't replace them, so at least not yet.

- To me that we believe this aspect of automation in vehicles is new. It's not. Everything from automatic transmissions to power steering are the foundations that this is built on. So we've been actually infusing automation into cars in a variety of different ways for decades, more and more, yeah. But the offshoot of that is the expectation that if you're gonna add more automation, folks are going pay attention the same way. They're not going to, why are we automating? We're automating to free resources. Now the question is as humans, how are we gonna behave? How are we gonna use those resources to do other things? And some of the data that we're beginning to see and understand is that humans are capable of interacting with the assistant systems in ways that we wouldn't have expected early on. So quite frankly, it's less about the automation, it's more about the interfaces and the support systems and how to connect us to that automation. Technology in and of itself is not going to solve a driver behavior problem. We keep adding more and more technology to cars with the hope and prayer that it reduces accidents, mitigates harm. The reality is we have to fit, we have to begin to sync technology focused very targetedly at the behaviors we wanna optimize. We, as humans, are not all above average drivers, although I'm sure many of us around the table here today and listening are thinking "Everybody's gotta be above average," but 50% is average.

- Yeah, absolutely. No, I think we can all be on one page that we don't like the Boston traffic and the many drivers that we encounter, I think everyone can agree with that.

- I'll take the Boston drivers over the New York drivers every day.

- See that's a hot debate topic, which city has the worst drivers? Which people often debate in the public sphere. Well, and public perception is a critical factor, perhaps the most critical factor in assisted and automated vehicle adoption. What does y'all's research suggest about the biggest barriers to public acceptance?

- One of the biggest barriers that's that's fairly well discussed openly is consumers understanding of what assisted and automated driving is in the United States. A recent piece out in the last couple weeks suggests that that's one of the bigger differentiating pieces between consumers in China and consumers in the States is that consumers in China really know what all this technology is, and are getting a much more mainstream understanding of that, so they're demanding it. And the US consumers don't really have clear understanding of what's assistive, what's automated, and the fact that full self-driving doesn't really exist. So when we think about that barrier, we really think about trust, trust in the technologies that's appropriately calibrated to the system's support for us, and ensuring that we're not overly trusting technology when we, as a human, would be a better driver, but we're leveraging that technology where the automation can support us most effectively.

- I can add to that also predictability. So people want to know what the system is gonna do, and they want to feel confident that it will behave the way they expect it to behave. So this is one another key thing, and another thing is we need to make automation more personal. So seeing in a video or reading about it is not the same as experiencing. So a short ride, 10 minutes in a Waymo car, for example, can do more to shift someone's perception than hours of media coverage. It's about helping people build the firsthand understanding. That's where real acceptance starts.

- And can you share for listeners who may not know what Waymo is?

- So Waymo is a Level 4 vehicle basically in certain areas in the United States, these vehicles are deployed, so basically you can schedule a ride with them and fully autonomous, there is no driver behind the wheel.

- Yeah, very scary for some people still, which is why we're here. But Bruce, what about for you in terms of what research suggests about the biggest barriers of public acceptance?

- One of the challenges that we've seen with the introduction of new technologies is how the consumer develops expectations about how a system will work, how do you actually use it? In fact, there's another aspect there of, "How do I use it? How do I turn it on?" So there's many technologies in the car that may be very well engineered in terms of their underlying function, but quite often there are challenges in terms of the user interface. Obviously, the user interfaces are developed by engineers who spent hours and hours thinking about and developing these interfaces, and they understand how the interface works in and out, right and left so that everything of course is, "Intuitive," quote, unquote, because they've been working with it for years. And it's that classic challenge just as you find in computer systems, quite often when the new feature comes out, it's intuitive to the designer because they've worked on designing it, but it may or may not be intuitive to the end user. If they are actually walked through it, they go, "Oh, okay, now I understand how this is meant to work." But unless we're given appropriate support to help walk us through this new technology, it's quite often the case that a number of people will never use it. And we also know very well that once a person gets, for example, a new car, if they don't try out a technology in the first three or four weeks, there's a very, very high probability they will never turn it on for the length of time that they own their car. And so just as it's a challenge with computer technologies and the like, in the vehicle, the whole issue designing user interfaces and finding ways to help talk the user through the feature so they can try it out and then intelligently decide whether or not they want to use it or not. And that human factors interface of helping introduce and explain features to get people to try them is really a key challenge for the industry.

- I think one thing to add to my colleagues here is that when you think about barriers here, we really have highly automated technologies on the road today in a number of cities. Ms. Pnina described most of these systems are supported extensively by teleoperation, someone in the back office whose helping these robots on the road. The barrier to public acceptance to me is how safe is safe enough? We have no yardstick on how safe assisted and automated technologies need to be to be accepted. And one of the things that I call on the regulatory side for is helping us form that social norm because a technology that can help improve safety by three or four times what it is today is something we probably should be endorsing. And the question is, is that safe enough tomorrow? And my answer is no, we need to be doing better tomorrow. But picking some reasonable targets and iterating from that and enhancing public acceptance over time. Otherwise, every incident there is, is a major news media circus, and that's really the key to avoiding, to enhancing public acceptance of this technology.

- Yeah, well, this is a great segue because speaking of regulation, how are government policy and regulation influencing Advanced Vehicle Technology development and adoption? You spoke to it a little bit, but if you could expand.

- So unfortunately, government policies today are not as effectively as non-government policies enhancing regulatory adoption of these technologies. So organizations like the Insurance Institute Highway Safety Consumer Reports, Euro NCAP, more non-government organizations that are either encouraging the adoption of these technologies a little faster. We saw late in the last administration reforms in the US NCAP program, which is really NITSA, National Highway Transportation Safety Administration, their effort to proactively promote these technologies, but it's a little later than optimal. So it's really about using the carrot and the education side of promoting technologies that we don't truly understand the safety benefit of. And that's where the white space I talked earlier that we discovered nearly a decade and a half ago was that many of these systems, well, we believe they're good, we don't know until the data is developed years later. So one of the pieces several years ago during the Obama administration was the collaborative agreement between the automakers, NITSA and IHS around the the deployment of automated emergency braking systems, which was really a Hail Mary pass at trying to accelerate these technologies on the road. So at the end of the day, we see the NGOs taking more of a leadership than the government here.

- This episode of "Supply Chain Frontiers" is brought to you by the MITx MicroMasters program in Supply Chain Management, a fully online program that offers access to top-tier supply chain management knowledge to anyone, anywhere with an internet connection. Visit micromasters.mit.edu/scm to see the courses that are currently running. Bryan, you mentioned this earlier, the cost of repairs as technology becomes more advanced in vehicles. One of the panels that interested me most at the AVT event was the collision repair panel, which was really meant to represent consumers, and they spoke to the high cost of repair as well as the insurance for vehicles with assisted and automated technology. What seems to be the consumer perception in this regard and how should industry respond?

- Consumer perception is that cars are repaired much like they used to be, and unfortunately they're not. These are highly complex technical systems integrating together a little sheet metal, a little aluminum, lots of different forms of plastic, sensing technologies, compute, wiring. These are among the most complex technologies that we as humans engage with on a daily basis. So when something happens, it's not going and just banging out the metal and reforming the structure of the car that way, it's about rebuilding the sensor architecture that was there to protect you in the case of an adverse event. So when we think about repair today, it's not auto body repair, what it used to be. It's bringing in a bunch of engineers to figure out, "How do we address this? Can we find the parts? Where might those parts be coming from?" And then quite frankly, calibrating the systems again, and hours and hours and investment to that. Things that take minutes when the car is flowing off the assembly line take hours to recreate later on. So when consumers think about risk and insurance costs, there's no mystery that the cost of ownership of a vehicle starts with your purchase if you can afford to purchase or lease a car, but the cost of insuring that and the lifecycle costs of that vehicle are driven by the complexities of sensing and compute technology. So I think one of the things that consumers really don't recognize enough is that the auto industry is advancing in material sciences at an accelerated pace. How do you best way to save gas is to make the vehicle lighter, so huge efforts in lightweighting. And the same is really there in a lot of other logistics avenues as well, whether that's long haul trucking or that's local delivery or robots chasing us around on streets to deliver packages in some cities today.

- Yeah, how do you think industry should respond given the higher costs now on the consumers?

- I think it starts with a strategic conversation, being willing to talk across the industry on how do we find new strategic pathways forward? Quite frankly, most of the organizations that I talk to know there is a cost problem. The question is how do you begin to address complex problems that require so many stakeholders involved? Each organization only owns a little piece of this complexity. So it's moving from a one industry talking to another or one partner talking to another to how do you get key players in a room to have a real big picture conversation, setting strategy forward where we're really thinking about repaving or restructuring how we look at the cost of ownership, the cost of repair of vehicles, maybe even the inspection and licensure of repair's facilities. The technologies required are far different. And this brings all topics such as right to repair, which has been hot in the state of Massachusetts over the years into play. Should a local repair shop be touching advanced software in the vehicle? Once someone does touch that advanced software, who should be liable for that? I mean, should a manufacturer be liable when you change the chip set in your car to something that it originally wasn't designed for? Where are the bounds of liability in? So these are huge conversations, but are critical if we're gonna get our hands around both repair and the long-term cost of ownership of vehicles. And I for one, believe we are going to be driving the vast majority of miles traveled in Western worlds mostly the old fashioned way with a little assistance involved for years, decades, if not the best part of a century or more.

- Okay, right because changing public perception takes some time to get everyone on the same page, of course.

- Just as a brief, practical follow up to that, one of the things that we've discovered as part of the AVT Consortium is unexpectedly, we've learned a lot about the complexity of repair. It's been a very positive aspect of our research study with our fleets of vehicles that we've instrumented and have people take out on the road. We've had no serious accidents in any of the vehicles we've had on the roads for the past 15 years. But recently we have had a couple minor, quote, unquote, "Fender benders," low-speed interactions, little bit of crumple on the rear bumper or the front bumper, and we've run into this firsthand. We've had the experience of taking the vehicle to the dealer where we got the vehicle from and literally taken two to three months to get the vehicle back on the road because the dealer in fact is not equipped to repair the sensors that are in the bumper anymore. More than that, the dealer is not equipped to calibrate the sensors that have been replaced in the vehicle. And if the main dealer that you purchased the vehicle from is not equipped for that, where do you go? So it's a very large and important topic that is quickly becoming very complex and more significant for everyone on a day by day basis.

- Well, and a very practical thing for industry to take up in terms of equipping their dealers with the skills to repair these things.

- And Mackenzie, that's an interesting piece is that where do the boundaries of what dealers are not owned by the manufacturers, they're independent. And this is a hot button issue with Tesla Service Centers, Rivian, the new Scout brand that's wholly owned by VW Group. So when we're thinking about software systems that are assisted or automating vehicle features, these are features that are going to be deployed in a new car and are today, but are gonna have to exist over the lifecycle of this vehicle. What do they look like 10, 20 years later when the windshield has begun to chip and the camera can't see a clear view anymore? Nobody knows. And these are some of the white space issues that I think are highly relevant to the future of safety that we don't even know how to effectively study now because quite frankly, we haven't seen cars on the road for long enough. And I think these are the big picture issues we need to be thinking about.

- Right, which speaks again to the importance of having AVT to have this long-term data right over the course of these developments. And looking at the impact that y'all's work has had on the industry, wondering if you could speak to some examples of how this data has informed industry practices.

- One of the things that does emerge from the data very early on is something as simple as video collection. It is totally eye-opening for engineers to go and look at these videos and see what people are actually really doing in the car. Some of the human behavior that you see, the engineer would never have considered needing to plan for. For example, we all have the image of somebody picking up and unfortunately trying to text while they drive. So there's some awareness of that, and you do things in terms of develop the driver monitoring system to hopefully counter that. But we've also seen things like people are driving down the road and not only picking up the phone to text, but taking large size iPads and setting them over the steering wheel and working with them while they're driving, and this is now covering up the airbag. And if the airbag was redeployed and pushed that iPad into the individual, that would be a really, really high-risk situation. So that's sort of one category is actually seeing what humans do. We do get a lot of feedback from our members, that particular analysis or this set of video or this sort of data has been really, really useful, but they don't always explain to us exactly how that has been because we're really getting into the area of proprietary development. So there are certain things that members will talk about together in the group. And actually that's worth emphasizing for a moment. One of the things that's been amazing for us is the AVT Consortium gets together multiple times during the year, bringing people together, and we get into active discussions of questions that are important to people today. And of course, nobody is compelled to speak up from any given company about what their concern is, issues that they're trying to address. But we have found that actually a number of companies are very willing to speak up and speak in very candid ways about the kinds of challenges they're confronting. And it's amazing what person from company A says, "This is a challenge we've been struggling with, and we have done X and Y to address it." And then you see across the room, person from company B says, "Well, that's really interesting. We've been working on that too, and we've tried X and Y and this is our thinking, but it would be great to dig in this deeper." And the amount of cross-industry exchange in these meetings has been a wonderful part of this whole process of bringing this Consortium together.

- And I'll echo Bruce's words, if you ask me what's the most impactful thing I think has come from AVT, it's the exchange of information between the partners. MIT in particular has been known for decades as a potential neutral space for industry to come talk together about research questions. I believe the only way we're going to compete in the automotive industry forward and compete with the growth of the Chinese auto industry is by partnering and accelerating innovation. Innovation in the US is perhaps our most critical commodity, and that's something I think MIT has done exceptionally well over the decades from the World Wide Web Consortia to many other efforts within CTL and across campus. And AVT is just another one in a long list of case studies where we are able to stimulate innovation across an industry.

- And as you both said, to give this forum that is not possible in the public space, but only can happen in anonymity. Let's dive deeper into the research, looking at the methods and the scientific process to get this opportunity that we have with you all to look and see what's happening behind the door. Curious if you could speak to some of the unique challenges in measuring driver trust and reliance on automation and how AVT ensures its research remains rigorous, while also being applicable to industry needs.

- So, trust is dynamic, right? It changes over time and depends heavily on the context. So it's not enough to just ask drivers, you know, once how much they trust the system. That is exactly what is unique about our approach is that we look at both what drivers say, but also what they actually do.

- Yes, as in all areas of life. Yes.

- So yeah, we use behavioral proxies for trust and reliance, things like how often drivers activate or deactivate a feature, how quickly they're reengaged when they are prompt to do so, or how frequently they glance at a system display. So we really look at how they interact with the system in the real world at scale. We track these behaviors over long periods across different road type situations, system updates. And that really helps us to understand not just one moment of trust, but how trust builds or breaks down over time. Of course, we strike a balance between real-world complexities and research structure. So as we mentioned multiple times in this discussion, we are collecting naturalistic data, but we ground our analysis in clear hypothesis, validated measures and replicable methods. This we can offer insights that are scientifically sound and directly useful for our industry partners in designing these systems.

- In looking at y'all's research impact, I know from our conversation there's more that you can't say than what you can in terms of specifics. But Bryan, if you could share what's an example of a key insight from your research that changed the way companies think about automation?

- Yeah, Mackenzie, I think it starts with a little bit of scientific data goes a long way. Most of the companies we work with in industry are often making good educated guesses based upon very impartial information because the development cycles are moving so quickly. So having data at their fingertips has helped a number of our partners accelerate innovation. As our colleagues at Toyota have discussed publicly, there's a number of datasets that have been created with them that have been used internally to R&D processes. But I think the most important impact comes from some research my colleague Pnina led here, that automation is not set and forget. The expectation with a lot of the assistive features earlier on was you're gonna turn this feature on at the beginning of the trip or the beginning of your time in the highway and you're gonna leave it on and it just space out. And quite frankly, that's not true. And some of the work that Pnina did with one of her postdocs really showed that we have a dynamic exchange going on here. At times I wanna drive, at times I'd love to be assisted by the automation. Well, I'm going to eat my cheeseburger on the way home today, and then turning on the automation may be a great thing at that point in time. So many of the systems like Autopilot assume that you're gonna have a hand on the wheel. They're meant to assist the driver, but the driver's supposed to keep their hand on the wheel. Other systems like GM Super Cruise, Ford BlueCruise are really designed to allow you to take your hands off the wheel. Not everybody agrees, but to me, designing for hands-free operation is critical. Why? Because the data clearly shows even if you develop a system to be hands-on, people are taking their hands off anyway. So if consumers are gonna take their hands off, we gotta design these systems for what they're being used as. And this really follows the safe systems approach that US DOT has been pushing over the last several years is that we need to consider how folks are using technologies and design technologies around how they're being used as opposed to a preconceived notion of how we'd like them to use them. And then again, this mirrors very much where AVT has been over a decade or more.

- Right, responding to human behavior as opposed to trying to dictate it. So looking forward for the future of AVT and what's next, where do you all see the future of vehicle technology heading in the next, say, 5-10 years?

- Well, I'll start. First of all, I believe that we will continue to see growth and partial automation and intelligent driver support systems. These technologies will get smarter, more adaptive, and more personalized, and better at recognizing when and how to engage with the human driver. At the same time, I do believe in Level 4 automation, and I think that we will see meaningful progress in that direction as well. But I also think that the road there is not just about refining the tech, it's about building systems that people trust, understand and are ready to coexist with. So my take on it is that in the next 5-10 years, we will see a future where Level 2 begins to scale in selected applications while the rest of the market benefits from better, smarter, safer and more intuitive support system that bridge this gap.

- So I see things in some ways very synergistically with Pnina, and in other ways a little divergently. I do agree that assistive driving features are a huge growth opportunity, whether it's Level 2 features or as the industry nomenclature goes, Level 2+++++. And the reason we keep adding pluses is that we want to keep the driver engaged and the driver responsible for the oversight of these technologies. I think ITS and connectivity services that are coming to the vehicle, either through the 5G network or other communicating with each other in the next 5-10 years is gonna begin to take off a little bit. Where I think that the automation sphere is heading is actually probably less is more for a while. I don't see the current growth in Robotaxis or highly automated trucking continuing. I think these are investment models that will begin to wind down over the next 5-10 years. While they're growing it, we believe huge scale now the economics don't make sense from a business perspective to me. I think the big change in automated driving to me is probably a little further out than the 5-10 year mark. It's probably more in the 2-3 decade mark. It's really highly pilot automation where I turn a highly automated driving system on, call it a Level 4 feature, and a vehicle takes over all responsibility for a period of time. And I can go for 2-3 hours or more where, or maybe not be able to fall asleep, I might have to stay awake and somewhat alert. But we begin to handle the complexities of these transitions, and the transitions being that the vehicle seeks safe harbor. If I don't intervene, say maybe that's pulling over the side of the road at a designated spot, pulling off the highway. But this is a feature that I think most of the auto industry is beginning to center on. Is it the long-term focus is an auto automation feature that can really relieve us. So lots of folks say, "Well, that's hard." Well, yeah, it is, but right now Robotaxis are supported, as I mentioned earlier, by teleoperation. You know, someone in the back room. The backroom operator here is not a fallback ready user that's there at any moment. But the backroom operator becomes the person in the operator's seat, not expected to take over instantly, but in the for of minutes or hours they're there on the spot. This is the feature that I think consumers are willing to pay a lot for, in essence giving me real functional time back, but leaving the bulk of the driving off the highway to the old-fashioned way. And why the highway? Because the highway is the simplest domain. So for the near future, things aren't gonna change a whole lot, a little more assistance and a little more safety.

- I think one of the key terms this has been used here is that of driver support. And I think that's really, really crucial in terms of both improving the functionality and improving the acceptance of the semi-automated systems that are in our vehicles now and are are being refined. One of the keys is that, as we've said, for Level 2 driving and the like, driver monitoring systems become increasingly important. The key is people are concerned with the whole idea of, "Oh my gosh, I'm being monitored." And there are a couple of important topics that have to be tackled there. One of them of course is on privacy. If this vehicle is going and monitoring me, is my privacy being invaded? Now, one of the things that most consumers don't realize is that most of these systems, and I will emphasize most, are not actually saving any video data. The video camera is sending in a stream of information that is processed to give it certain information, where a person is looking, are they looking, et cetera. And that coded information is being utilized, but the video image itself is not being saved. So vehicle A doesn't have a record of, "Oh my gosh, the argument I had with my wife." And so it's one important for people to know that. At the same time, there is a real role for the government or other organizations to really address this issue that in fact that should be the standard. Because in fact, there are other companies to our understanding where in some cases the video is being saved, and that could be argued as inappropriate. We really do need to have a reasonable expectation of privacy. Another kind of support comes to, "Well, what does the driving monitoring system actually do with information?" And not in the privacy sense, but in terms of communicating with the driver. There's been a historical emphasis on introducing warning systems into the vehicle. So a lot of the monitoring systems at the moment are really emphasizing on beeping and booping at me when something is inappropriate or wrong or I'm not paying attention or whatever. The problem is, there are so many systems in the car now when things go beep and boop, it's like, "Where is that coming from? You know, what thing is it trying to communicate?" And we need to get out of the mode of just always warning to beginning to provide more of an information flow so we are giving information back to the driver to help support them in the driving process. So in fact, if I have looked off the road for too long because I've gotten absorbed in something, having a tone that has a reasonable expectation to catch my attention is one thing. And if it's designed in a way that I see that as a support that, "Oh, that helped me get my attention back to the road," that is gonna be both more functional and more acceptable as opposed to, "Oh my gosh, the car is buzzing at me again." So I think that's a really important aspect of moving from alarm to support.

- Yeah, a great point, right? Because people begin to also tune out those warnings. Bruce, you spoke to how do we solve for a lack of driver engagement and situational awareness a bit with a different approach to the warning systems. Pnina, what might you say in terms of some solutions to solve for responding to a lack of driver engagement and situational awareness, particularly with assisted and automated vehicles?

- So yeah, so one of the biggest challenges in designing systems that keep drivers appropriately engaged with overwhelming or distracting them is exactly that, keeping them engaged. Drivers are actively learning how these technology behave and in some cases learn how to work around them so that creates a moving target for designers, basically to find a way to create systems that are robust enough. I do see this as an opportunity with all the, especially AI applications that are being introduced, specifically large language models that starts to enter into the vehicle environment. These systems open the door for more natural interactions that could help keep drivers informed and engage in meaningful ways. So the sweet spot is a shared control, where the vehicle supports the driver, provides clear feedback and adapts to the condition or as the condition requires without taking the human out of the loop.

- Yeah, absolutely. So you all have reached your 10-year milestone, what's next for the Consortium in terms of research focus or new initiatives?

- Well, as we move forward to our second decade, we are expanding our focus to few key new areas. One is understanding how people adapt to increasingly intelligent systems, whether that's automation, electrification, AI-based assistance in the vehicles. Other directions are focusing on personalization.

- I think that also one of the intriguing things about the size of the dataset that we've been developing over the past 10 years, one is that we are always going and adding new vehicles to the fleet to go and specifically look at how new implementations of systems have impact on driver behavior and overall apparent safety. So continuing to add vehicles is an ongoing part of the fundamental process. At the same time, as the overall database grows, we have the opportunity, which we've been taking advantage of, the increasing computer power modeling approaches, large language model techniques, et cetera, we always have a continuing opportunity as new ways of studying data and analyzing data are developed, is going back and using these new analytic techniques on the existing dataset. So not only can we continuously go and begin to address new questions, we can also use new tools and apply them to the large set of data that is always growing.

- Tomorrow is old news a year or two from now. And as we look forward, as Pnina mentioned, into our second decade, it's really continuing to evolve in looking at the intersection of behavior, technology and policy in new ways. And we're very focused on what are the new white spaces where we can enhance and find safety gaps that exist and help fill those ourselves and with our partners involved very much in the CTL model of partnership with the organizations that support and sponsor our work.

- That wraps up this episode of "Supply Chain Frontiers." A big thank you to Bryan Reimer, Pnina Gershon, Bruce Mehler for sharing their expertise and insights into the evolving landscape of vehicle technology and its impact on supply chains and mobility. To learn more about the AVT Consortium and the research discussed today, visit avt.mit.edu. "Supply Chain Frontiers" is recorded on the MIT Campus in Cambridge, Massachusetts. Our sound editors are Dave Lashinsky and Danielle Simpson at David Benjamin Sound. And our audio engineer today is Kurt Schneider of MIT Audio-Visual Services. Our producer is myself, Mackenzie Berry. Be sure to check out previous episodes of "Supply Chain Frontiers" at ctl.mit.edu/podcasts or search for us on your preferred podcast platform. If you enjoyed this episode, please subscribe. I'm Mackenzie Berry, thanks for listening, and we'll catch you next time on "Supply Chain Frontiers."