Episode Preview with TTI's Bob Brydia (audio, 38s):
Full Episode (audio):
March 30, 2022Episode 30. Two Steps Forward, One Step Back? Keeping tabs on self-driving tech.
FEATURING: Bob Brydia
One of the first lessons we learned about autonomous travel remains true today: Building a self-driving car is a lot more difficult than many people expected. Senior Research Scientist Bob Brydia sits down with us again to discuss progress made in the past year related to self-driving vehicles becoming commonplace on our roadways, and how far they have yet to go.
About Our Guest
Senior Research Scientist
Bob Brydia is a transportation technologist with more than 30 years of diverse experience with national, state and local transportation research sponsors. Bob has led several large-scale deployment projects, including testing automated vehicles in real-world environments, and has developed multiple innovations to advance transportation.
Bernie Fette (host) (00:14):
Hello, and welcome. This is Thinking Transportation — conversations about how we get ourselves and the things we need from one place to another. I’m Bernie Fette with the Texas A&M Transportation Institute.
Bernie Fette (00:31):
When we started this podcast a little more than a year ago, autonomous travel was one of the first topics that we explored. We talked at that time about out how the lofty promise of self-driving cars was not lining up with the harder reality that manufacturers faced in trying to build them. Bob Brydia, a senior research scientist at TTI, was with us then to explain some of the reasons for that disconnect. We’ve invited him to return to help us understand where progress has been made, and where obstacles still remain. Bob, welcome back.
Bob Brydia (guest) (01:10):
Thanks, Bernie. Pleasure to be here with you.
Bernie Fette (01:13):
It’s been just over a year since we had that conversation about the progress toward a future of autonomous vehicles. And I recall asking you at the time, why self-driving cars weren’t all over the roadways yet, despite industry promises that they would be, and you told us that there was a very straightforward reason why. Because building those cars is simply a lot harder than people thought it was going to be. So if I ask you the same question today, would your answer be pretty much the same, too?
Bob Brydia (01:49):
It would. And as a bunch of different companies have admitted in the press over the last year or so, they are finding that it’s harder than they anticipated it to be, as well.
Bernie Fette (02:03):
Please tell us why.
Bob Brydia (02:05):
Well, I think we discussed last time that there have been a lot of thoughts expressed that all we have to do is visualize the road ahead, infer what’s on the road and we get to drive down the road. What that’s forgetting is all of the edge cases. So the road that curves in a manner that’s unexpected, the pedestrians walking out in front of the vehicle, work zones, weather conditions that cameras and sensors can’t see through, such as snow on the road, that obscure lane lines. All of those types of things make this an extremely challenging environment. And over the past year, I would think a number of companies have recognized that it’s harder than they initially stated in their press statements.
Bernie Fette (02:53):
So maybe this is a good time to take a fresh look at where you’ve seen some progress. For instance, I think we about this once before too, you can get your pizza delivered that way, at least in some places now, I think. Even if it’s just pilot projects, where else have we seen progress? Can you just tell us the main areas and talk a little about each?
Bob Brydia (03:14):
Sure. I think the main area of progress that everybody has seen in the last year is with heavy vehicles, you know, class eight semi tractor trailer trucks driving along the roadway. Many of those companies have had a great deal of success, navigating freeway conditions and doing so very successfully and without crashes. Those companies are moving on to what is called driver out scenarios, where there’s no longer a safety operator safety driver in the vehicle itself. There was one done by a company that was approximately an 80-mile trip. It took place at night and it was both preceded and followed by essentially chase vehicles and vehicles looking out for situations that could affect the vehicle and its operations and ensuring that it was safe.
Bernie Fette (04:05):
And that example was, I guess, out on an open road that was far removed from population centers?
Bob Brydia (04:11):
That was on a freeway between two cities.
Bernie Fette (04:14):
Which makes me wonder is that one of the reasons why there’s more progress seen is because you’ve got some less complicated, less crowded driving environment out there. If you’re testing a truck like that between two cities out in the rural area?
Bob Brydia (04:29):
Definitely the concept of freeway driving between origin and the destination point is probably the scenario where there’s the most amount of information known, the least amount of variability, and the best opportunity to perform a trip safely without running into significant unknown conditions. There are a number of states, Texas included, where there are several companies operating autonomous vehicles on freeways, performing actual freight operations for paying customers. So those operations are at the beginning stages of moving the industry forward with autonomous trucking. But at the same time, they’re proving that the concept is feasible. The challenges still exist at endpoints and unexpected conditions. You don’t want to try to take a tractor trailer and, uh, navigate it autonomously through downtown city streets, making turns and dealing with pedestrians and everything. But on a freeway, those situations obviously do not apply.
Bernie Fette (05:40):
We’ve heard before about how last mile, just from a shipping perspective, that there are challenges involved in that last mile of a journey that may have been well over a hundred miles, but that was for reasons that are different from the ones that you’re talking about now.
Bob Brydia (05:59):
Well, first mile and last mile, when you’re talking about autonomous operations really are focusing on the aspect of the vehicles going through more crowded conditions, more significant potential for events that are not experienced often by the computer. So those are the challenges. What’s coming up now in terms of thought leadership is that perhaps we have transfer points. So you have an actual human that gets into the vehicle, drives it from, say a port or an inland port to one of these transfer points right off the highway, autonomous operations, then take over, drive it on the highway to very near its destination, location, goes into another transfer point. Then a human operator takes over. And so on both sides of the trip, you have a human in the vehicle driving the vehicle. And in between you have autonomous operations.
Bernie Fette (06:59):
I think you also mentioned something about how self-driving taxis, how that was another area where you’ve seen some progress being made. Can you elaborate on that a bit?
Bob Brydia (07:10):
Sure. So in the self-driving taxi arena, taxi companies within cities are just now entering essentially a pilot phase. In the past year, there have been a number of opportunities for companies to take passengers on paying trips in driverless vehicles, internal to the city. Now, typically those are lower speed trips. You know, we’re not talking 70, 80 miles an hour on a freeway. And we’re talking about, uh, vehicles that are very heavily monitored by an operations center and on routes that have been very diligently mapped to ensure that anything along the route that would cause significant events or conditions are already known and don’t surprise the vehicle. There’s a lot more to getting the public used to and willing to participate in those types of trips with vehicles that don’t have drivers in them. There’s the cost associated with it. Overall, it’s just a more challenging environment from the societal aspect, the economic aspect and the actual physical environment. But there is progress being made in that front as well.
Bernie Fette (08:27):
Self-driving taxis are basically the same size vehicles as the privately owned sedans that people might purchase. So I’m wondering why progress with taxis, but not so much progress with the privately owned self-driving cars.
Bob Brydia (08:42):
That’s a really good question, Bernie. Part of it comes down to cost. Those vehicles are still extraordinarily expensive to equip with all of the sensors and processing power that they need to infer the environment from the input that they receive. That’s not a cost that the typical consumer could even come close to affording at the present time. Also, that technology is not nearly as mature as the technology in the more refined environment of a freeway. So there are a lot more conditions that take place on city streets — work zones, pedestrians, bicyclists, scooters — all of those different types of situations occur way more often on a city street than they do on a freeway route.
Bernie Fette (09:34):
You’ve helped us understand about where the progress is being made, Bob. Can you talk a little about the obstacles that remain? For instance, you mentioned to me recently, something about how workforce issues are emerging as a concern. Can you tell us how?
Bob Brydia (09:51):
Sure. So before I get to workforce, let me just touch again on what’s called all the edge cases. So there are all kinds of conditions that happen in the daily commutes of everybody that drives behind the vehicle where your brain automatically just recognizes what’s going on and adapts, uh, essentially instantaneously. The challenge in an autonomous vehicle is to get them to experience enough of those types of situations, where it can do the same thing. So that’s one of the biggest challenges. When you look at things like snow and ice covering the lane lines on the roadway, work zones that appear unexpectedly, or work zones where the lane lines have been completely obliterated. And so you’re getting discrepancies in terms of where the barrels are telling you to go versus where the lane lines are telling you to go. All of those are physical challenges that sensors have to navigate.
Bob Brydia (10:51):
There is however, as you said, Bernie, the workforce issues associated with autonomy. In essence, we have a nationwide shortage of truck drivers and have for several years, the concern is that autonomous operations will wipe out an entire class of jobs and a portion of the economy. And you have two sides of that equation. You have the autonomous company saying that’s not gonna happen. We have a shortage now. We’re just trying to make a little bit of a dent in that. And we’re still gonna have operators in the trucks for a long time, even if it’s at those initial transfer points to and from. You also have the organizations that represent truck drivers that are looking at some studies that say that, you know, autonomous operations can completely wipe out that portion of the workforce. Given the fact that we have shortages of hundreds of thousands of drivers right now, that doesn’t seem to be very likely, at least not in the near future.
Bernie Fette (11:55):
Okay. If you had a research sponsor right now hand you a really big check to study, to explore whatever self-driving issue you thought was important, whether it’s augmenting work that might be underway elsewhere or a completely new idea, what would you do with that research funding? What question would you set out to answer?
Bob Brydia (12:20):
It’s a really interesting question to consider. If there are any sponsors out there, would love to talk. However, uh, I think we’d focus on the human aspects of things right now. We’d focus a lot on the trust aspect of having these vehicles perform these types of situations, not only from the standpoint of a consumer or customer getting into a vehicle and telling the vehicle to drive to a location, but also from the standpoint you mentioned before — delivery robots and getting your pizza delivered to you, those trials are taking place in multiple cities and things. Now let’s look at how the public really accepts that. How does the public accept the fact that the tractor trailer next to them and the four behind that one don’t have a human in them? How do we deal with pulling over a vehicle that is autonomous, that has a safety problem, or that needs to pull over because of an incident or something along those lines? That involves the whole societal aspect of putting these vehicles on the roadway beyond just the technology. The technology’s developing, the software is developing, but we still have a long ways to go on the societal aspect of accepting this type of environment on our roadways.
Bernie Fette (13:41):
I think the phrase that we’ve heard in the past is early adopter. Those people who are standing out in front of the stores whenever the latest iPhone comes out. And then there are those who may not fall into that category. I guess what I hear you saying is that same is true for this type of technology, too.
Bob Brydia (14:00):
Early adoption is certainly one critical aspect of understanding how society is going to view these vehicles. There’s also economic opportunities that need to be considered. For example, in freight. Freight is a very competitive business with possibly pennies per mile of profit. So anything that you can do to increase the efficiency of an overall route and do that repeatedly might make a very significant difference in the bottom line profit of a company. So certainly those are going to drive some of the progress forward in these types of fronts.
Bernie Fette (14:41):
Let’s go back for just a minute to one of the original promises that was made that we were reading about years ago. From the very beginning, companies designing and building autonomous cars have been maintaining that they will eventually eliminate most of the crashes that injure and kill people. To do that, as we had talked about when we had this conversation about a year ago, the machines have a lot of learning to do. How does machine learning work? Can you explain it for the non-engineers among us?
Bob Brydia (15:17):
So machine learning is nothing other than a data analytics technique to allow a computer to do what humans do by nature, which is learn from experience. So machine learning takes repetitions — millions and millions of repetitions — and teaches the computer to learn by itself what’s going to happen.
Bernie Fette (15:42):
So is it accurate then to say that to at least some extent, machines learn in a way that’s at least somewhat similar to how humans learn?
Bob Brydia (15:53):
Well that’s the holy grail, right? The holy grail of artificial intelligence is to create a computer that can think the way a human brain does. Now they’re doing it by one aspect of artificial intelligence, as you said, called machine learning, which is essentially brute force repetition and, uh, pattern recognition. So if you recognize that pattern enough times the next time it comes up in a situation, you’re gonna say, aha, that’s something that I’ve seen 10,000 times before, and this is what I needed to do. In true artificial intelligence, a human brain may be able to look at that and infer what they need to do, even though they haven’t been in that situation before.
Bernie Fette (16:43):
Which would not be what you described a moment ago as brute force learning. Right?
Bob Brydia (16:48):
That is correct.
Bernie Fette (16:50):
So you’ve got basically those two different forms and one of them is a little more similar to human learning than the other.
Bob Brydia (16:58):
Yeah. So artificial intelligence is the overarching area. And as I said, the holy grail is to have a computer that thinks, operates and goes through life as a human does thinking its way through situations. There are many examples that you could pull from movies. Terminator comes to mind as to, you know, a fully operating machine that is able to think on the fly and deal with certain situations. Machine learning is a subset of artificial intelligence that again, focuses on that brute force pattern recognition of events that have been seen previously, so that if they’re encountered again, they know what to do.
Bernie Fette (17:40):
And then continuing on the safety aspect, what this part of our conversation makes me wonder is, is it possible for machines to learn bad habits just as humans do?
Bob Brydia (17:53):
That’s interesting because there are recent news articles about some autonomous vehicles that have not obeyed traffic laws. That’s likely the result of programming as opposed to machine learning. But nonetheless, the general answer to the question is yes, machines can learn bad habits, whether it’s through incorrect pattern analysis or through programming that didn’t take into consideration certain aspects of the built environment like laws.
Bernie Fette (18:31):
Sounds like this is really getting to the heart of your original point earlier in this conversation and a year ago. This is a lot harder than some people may have expected it to be.
Bob Brydia (18:41):
Definitely. And a number of CEOs of major companies have come out in the last year or so, and said that while they had made original predictions of self-driving fleets by 2016, 2017, early 2020, they’re essentially now refraining. Most of them are essentially now refraining from stating when they’re going to have these types of vehicles out there in full fleets, in operating on the roadways with everybody else because it’s unknown. And an additional challenge to that is the government regulations haven’t caught up with the pace of technology. And so right now companies have to petition the federal government to allow vehicles on the road that don’t have a steering wheel and don’t have brakes and gas pedals and rear view mirrors and all of those types of things, which are mandated in vehicles today. And that legislation is sitting in Congress, hasn’t been acted on yet. And so, uh, it’s a little bit of a, uh, wild west out there in that essentially, all 50 states have 50 different sets of laws that govern this and there’s no uniformity across the nation. That makes it extremely challenging for companies to have a operating environment that they can focus on consistently.
Bernie Fette (20:05):
And we’ve got a historical point of reference too, I guess, in that more than a hundred years ago, whenever the first automobiles started to become a little more common, the cars came first and the laws and the rules of the road came later.
Bob Brydia (20:20):
That’s correct. And there were some very interesting laws, such as a horseless carriage had to be preceded by a person holding two lanterns at a distance of at least 50 feet ahead. And some of those laws were on the books until the mid-1900s.
Bernie Fette (20:36):
Pioneering in traffic safety.
Bob Brydia (20:38):
Bernie Fette (20:40):
Okay. You’ve told us that the CEOs of these companies have stopped making predictions about when we’re going to see self-driving cars in greater numbers out on the roadways. But you’re not a CEO, which means that you can probably get away with making that prediction. What do you think?
Bob Brydia (20:56):
I’m gonna give you the same answer that I did a year ago, Bernie. We’ve seen a lot of progress, but I don’t think we’ve seen enough progress to state that, oh, they’re gonna be here in huge mass quantities within five years or so. I still think that we are two to three decades out from seeing autonomous vehicles as a common, everyday occurrence across all aspects of the traffic stream. There’s a lot of reasons for that — cost technology, learning, curves, public acceptance, uh, everything that goes into that statement.
Bernie Fette (21:31):
On the subject of learning. I’d like to wrap up with a question about your relatively new role. In addition to being a full-time researcher, you’re also an adjunct professor of practice in the department of industrial and systems engineering at Texas A&M. What’s the most compelling piece of advice that you would give to a young person who’s just getting started on a career path that’s similar to yours.
Bob Brydia (21:59):
I think the thing that’s been most valuable to my career is curiosity. Other people may see it as I never say no, but I’m intensely curious about virtually everything that goes on in this profession. So channel that curiosity that you have about whatever your profession is and utilize that to remain interested, remain focused, continue to move forward with the times as technology changes, as society changes. I think that’s how you stay interested and motivated and engaged throughout an entire career.
Bernie Fette (22:39):
Bob Brydia, senior research scientist at TTI, and now officially a veteran guest of this podcast. We’re glad you could join us again, Bob. Thank you very much.
Bob Brydia (22:50):
Thanks, Bernie. I enjoy being here and I enjoy talking to you on these subjects. Always interesting.
Bernie Fette (22:57):
Humans by nature learn from experience. Computers learn that way too. But it may take millions of repetitions for a machine to know exactly how to respond in a given situation. That’s certainly true when the situation involves driving a car or truck. When we consider all the potential surprises that result from jaywalkers, weather, scooters and other unknowns, the possibilities for dangerous conflict are virtually endless. Consequently, stakeholders across the board are learning an important lesson of their own. Teaching a car how to drive itself is a lot more difficult than many of us originally thought it would be.
Bernie Fette (23:42):
Thanks for listening. Please join us for our next episode — a lesson in the evolution of roadside safety. Crashes can happen despite our best efforts. But as Lance Bullard will help us understand, there are ways to make those crashes more survivable. That’s something he and his team have been doing for decades. Thinking Transportation is a production of the Texas A&M Transportation Institute, a member of the Texas A&M University System. The show is edited and produced by Chris Pourteau. I’m your writer and host, Bernie Fette. Thanks again for listening. We’ll see you next time.