If you’re wondering when we’ll see autonomous vehicles hitting our nation’s highways, don’t hold your breath. At least that’s the word of Dr. Phil Koopman, who has been following the development of autonomous vehicle technologies for more than 25 years.
Associate Professor at Carnegie Mellon University (CMU) department of electrical and computer engineering, Koopman is a leading expert on driverless vehicles and the security systems they require. He has extensive experience in software security and quality, and as a faculty member at CMU, he teaches young engineers the software skills necessary for mission-critical systems. He recently spoke to DC speed Group Editorial Director David Maloney on the outlook for autonomous vehicle technologies and the challenges they present.
Q: You have been working on autonomous vehicle technologies for over 25 years. Why has it taken so long for these technologies to reach their potential?
A: I started getting involved in the mid-1990s. I worked with the navigation lab team at Carnegie Mellon University, who hired me as their safety officer. Right before they took me on they’d been coast to coast [with an autonomous vehicle], from Washington, DC, to San Diego. It was 98% without the steering wheel. Think about it: 98% hands-free, coast to coast in 1995! I got hired because after that trip, they said, “You know, maybe we need a security guard.”
And then, I think it was 1998, there was a demonstration of an automated freeway system, where they shut down part of the freeway near San Diego and drove a bunch of platoon cars and a city bus on the road – no hands on the wheel on a closed stretch of the Interstate Highway. Again, almost 25 years ago, right? So the way I like to look at it is that we were 98% self-sufficient across the country in 1995, and we’ve been working on that last 2% ever since.
Q: Why did it take so long?
A: Well, the catch is that the last 2% is really tough, and that’s fundamental to the problems in this industry. You can have a vehicle that’s good at the easy stuff – and that’s certainly an impressive achievement – but we’ve been around since the 1990s in a sense. It’s really that last 2% that’s hard because it’s always something new. It’s something you’ve never seen before. There is an endless variety of weird things in the world, and dealing with it all turns out to be much harder than people would like.
Q: Many driver assistance technologies are available on our cars today: lane departure, automatic braking. Even my Toyota can pretty much park itself. Are these simply steps towards autonomy?
A: They are a contribution. In reality, what was happening in the 90s was more like that than full autonomy. It was automatic lane keeping and things we today call driver assistance. These are important to have, but making them better and better doesn’t solve the autonomy problem. The reason you need a human operator in driver assistance vehicles is because the machine learning part is good at knowing what it knows, but it’s really, really “fragile” for things that you don’t know. she has never seen before. That’s the point of having a human driver – to deal with what he’s never seen before.
Q: So when will we realistically see self-driving trucks on our highways?
A: It’s more a question of How? ‘Or’ What than when. If you want to completely replace a trucker, that’s far in the distant future — and by the way, truckers do more than drive. I don’t have to tell your audience. But even for the driving part, it’s far if you don’t want to limit what happens.
If you’re willing to do something like take a stretch of interstate highway, and every day someone comes by and makes sure all the lane markers are there and there’s been no paint or oil spillage to hide the lane markers and there is no big pile of sand and there was no landslide and everything is fine – if you are willing and can -be there’s a guide vehicle that the automated trucks all follow in a conga line and the guide vehicle is responsible for making sure that if there’s an animal in the road it’s scared—if you are ready to make these kinds of concessions, it could happen in the next few years. But I don’t see anyone next year just saying, “OK, here’s a thousand trucks. Let it rip! I don’t see it coming as soon as a lot of people say it.
Q: Do you see this as the next step – where you would have a lead vehicle with a driver followed by a platoon of self-driving trucks?
A: I think a guide vehicle makes a lot more sense than having every truck do everything next year. But I don’t see anyone trying to market that.
Q: Do you think there will be dedicated lanes for self-driving trucks, or even dedicated highways?
A: It’s really hard to know how it will turn out. There’s a tradeoff between how much infrastructure you want and how difficult it is to get the vehicle to do everything a human driver would. I think it makes perfect sense to do things like have dedicated on and off ramps at logistics hubs. There may be a HOV lane. It will depend on the route. It will depend on the conditions.
Another way is to have designated times of day — times when traffic is light — for self-driving trucks to use the highway. The more you have the road to yourself, the easier it is to ensure safety.
Q: You and I both live in Pittsburgh, where they’ve been testing self-driving vehicles on city streets for several years. But wouldn’t it be easier to test the technology on highways and limited-access highways than in urban settings?
A: Well, let’s see it all. Right now in Arizona, Waymo actually operates driverless robo-taxis in a very, very benign environment.
The problem with city roads versus freeways isn’t that one is easier; is that the challenges are different. In urban environments you have a lot of crazy things going on all the time. It can be a very chaotic environment, depending on where you’re driving, but the good news is that if you drive slow enough, you can often lock up the brakes and be fine. You just have to know when to block them.
If you’re on a truck that goes 60, 70, or 80 miles per hour, however, locking up the brakes isn’t a very attractive option because you have a lot of weight and mass – and a lot of momentum. So it’s less about the weird and chaotic things, like pedestrians jumping off the sidewalks in front of you, and a whole lot more about planning ahead. I would say that intuitively it looks like the freeways are easier, but it’s not that easy; it’s just that the problems are different.
Q: We have seen many advances in machine learning and artificial intelligence over the past few years. What role will these technologies play in the development of autonomous vehicles?
A: Artificial intelligence means what is really hard to do. And every 10 years its meaning changes because some things get easy, while the next one gets hard. People tend to use “AI” as a catch-all term for all new technology, but it really doesn’t mean anything.
Machine learning, on the other hand, is a very specific technique. In machine learning, you show the computer system a bunch of examples and it performs statistical analysis. And then if it sees a new sample, it will compare this new sample with the original sample. So if he sees a person, he doesn’t really know it’s a person. He says, “You know, this thing looks a lot like every other person I’ve seen before, so it must be a person.” And that’s great.
If you train him on things he has seen, it works great. But it’s like 98% or 99%. If there’s something he’s never seen before, not only does he struggle, he doesn’t even know he doesn’t know what’s going on.
For example, there was a case where the system had trouble seeing people wearing yellow. It turns out that this system wasn’t trained to recognize someone in yellow, and so if you were wearing yellow, you were basically camouflaged by the machine learning system, which isn’t so great if you’re directing traffic to a construction site or you’re a cyclist in a yellow raincoat.
So the instances where he makes weird, crazy, or stupid mistakes is sometimes a real surprise to people, and that’s why I was talking about the “long tail,” the few things you haven’t seen before. That’s why everything takes longer than everyone wants.
Q: In the logistics market, the use of autonomous vehicles has obvious advantages, such as reducing the shortage of truck drivers. There are other benefits too: trucks can be spaced further apart, which could help reduce congestion on our highways, and driverless trucks might be able to operate for longer periods of time if they are exempt from driver hours of service regulations. . What are the key benefits that will help drive this technology forward in the coming years?
A: Well, let me come back to the question of jobs because it’s so central. If someone is a truck driver today, I don’t think they should worry about losing their job before they’re ready to retire. This technology will take time to establish itself.
And even if there are a thousand [autonomous] trucks on the road in the next four or five years, that’s just a drop in the bucket. It will take a long time to scale this technology to be able to take roads that are not the easiest and most benign roads. So the scary headlines about truckers out of work next year – it’s just not going to happen. On the other hand, I think the prospect of finding relief from the pilot shortage is fantastic.
As for the other things, all the things you mentioned depend on security. Until we can provide security, none of these good things will happen. And the industry is at a point where it’s only just beginning to think seriously about security.
There’s a saying we have in the IT world that the first 90% of the project takes the first 90% of the time. And the last 10% of the project takes up the remaining 90% of the time. In the end, it boils down to: Can you really ensure that these things will be at least as safe as a human driver? We’re not at the point yet where we have an answer to that, so there’s still some work to be done.