PROFESSOR FRAZZOLI, ARE SELF-DRIVING VEHICLES WITHIN REACH OR STILL A FAR-OUT VISION?
It’s a question of defining autonomous driving. Generally speaking, it’s happening. The technology is advancing rapidly and in some instances already pretty mature. On the other hand, I think we are not ready in terms of regulations as well as political and legal aspects. How do you get a license for such a vehicle, how to insure it?
AND WHO WILL BE ABLE TO USE THEM – A FEW SELECT RESEARCHERS AND SOME WELL HEELED PEOPLE?
We have to talk about affordability. A lot of people are working on the technology we need for autonomous driving, including myself. Many of the hardware and software components are still too expensive. They are plainly and simply unaffordable for normal consumers. If I had that much money to spend on a car, I’d buy a great sports car and drive it myself.
SERIOUSLY, WON’T THE COSTS FOR THE NECESSARY TECHNOLOGY COME DOWN VERY RAPIDLY?
That trend is clearly visible. Video sensors become cheaper and better all the time, and they provide a lot of information. Laser scanners are very convenient, but they are expensive and I wonder if you really need them. The challenge is to figure out how to reduce the sensor package on a car, make it smaller and cheaper.
CAN YOU FORESEE A STANDARD EQUIPMENT FOR AN AUTONOMOUS VEHICLE?
I would rephrase the question. What do we mean when we speak about autonomous driving, and what kind of hardware and software do we need to get there? Take the DARPA Grand Challenge which was held in 2004 and 2005 and thus almost a decade ago. Its aim was to master completely autonomous driving. All cars had to collect and process the information only using their onboard systems and sensors. That makes sense when you look at the fact that DARPA (Defense Advanced Research Projects Agency) is primarily interested in military applications, e.g., operating autonomous vehicles in enemy territory. But this concept doesn’t make a lot of sense for driving in a city with several hundred thousand or even millions of other cars. We have a lot of prior knowledge and infrastructure we can use to reduce costs and the complexity of the task.
WHAT EXACTLY DO YOU MEAN BY PRIOR KNOWLEDGE?
Data that’s already out there. Take something like Google Streetview and add to it vehicle-to-vehicle or vehicle-to-infrastructure communications. That way, cars can communicate wirelessly with the roadside, with traffic signs and lights, with traffic sensors, even traffic cameras and other vehicles. All that richness of data makes the goal of achieving autonomous driving much easier. A car can use the environment and other cars as its eyes and ears. The DARPA Urban Challenge 2007 was a step in that direction, but by focusing more on urban traffic it also shows where many problems still lie. Cars could recognize urban elements like another car at an intersection. But you still have to know: is that car’s engine on, is it moving, is somebody sitting inside? All that is very hard to answer with just on-board sensors.
YOU WERE TALKING ABOUT THE QUESTION OF DEFINING AUTONOMOUS DRIVING. CAN YOU GIVE US YOUR DEFINITION?
Very often, people think it means replacing the human driver. That’s nice and good, but I don’t think it’s necessary. Also, do we really think humans drive so well that the goal is for technology to replace a human behind the wheel? I don’t think we should take human drivers as the paradigm. Autonomous vehicles can drive very differently than a human would, as long as they do it safely and in an environmentally friendly way. I propose we think of autonomous vehicles as a subset of all the possible driving scenarios because that’s much more useful.
WHAT ARE, TO YOUR MIND, THE REASONS WHY AUTONOMOUS DRIVING IS USEFUL?
If you listen to the debate, the advantages of automated driving generally fall into three categories. First, Safety, because we can reduce or remove risk factors due to human error, coupled with mobility options for the disabled or the elderly. Second, the efficiency of having features that let you drive in a caravan on a highway and using adaptive cruise control. And third the environmental impact. Using these technologies, we can reduce carbon emissions anywhere from 20 to 50 percent. But there’s one thing most people forget: The above arguments lead to an improvement over the current status of the automotive industry, but do not change the basic concept of how we think of cars.
The ‘sharing economy’ and autonomous driving are two sides of the same coin.
HOW SHOULD THE DEBATE ABOUT AUTONOMOUS DRIVING BE FRAMED THEN?
It should be about how it allows us to fundamentally rethink transportation. We only use our cars 5 to 10 percent of the time, the rest of the time they are sitting somewhere, often a very expensive piece or real estate. That’s not a sustainable model, so I think the idea of sharing goes hand in hand with autonomous vehicles.
CAN YOU ELABORATE?
Take today’s carsharing services. They hold much promise, but have a lot of room for improvement when it comes to quality of service and availability. Now imagine, these cars could park themselves, go to gas or charge stations by themselves or reposition themselves to another location based on where a customer is waiting or demand is predicted. That’s much easier than a general-purpose autonomous vehicle, because those would be cars moving autonomously only along pre-defined routes. Moreover, since there are no human passengers on board to please, the cars could drive in a way that maximizes safety, rather than passenger comfort or travel time.
THEY’D STILL HAVE TO NAVIGATE REGULAR URBAN TRAFFIC WITH OTHER CARS, CYCLISTS AND PEDESTRIANS.
Absolutely, but the number of routes they need to know drops dramatically. They don’t have to drive the fastest route. They could well take the easiest and least congested route. They could have a yellow light on the roof as trash and other public service vehicles do to let others know they are coming. Things could be easier if you separated human drivers and autonomous vehicles, even on separate lanes. They could just drive slowly on the roadside, like service vehicles do. Human drivers already accept that today. I work in aeronautics, and the concept I like to think of is the “flight envelope” – what a plane can and should do in a given environment.
HOW DOES THAT CONCEPT APPLY TO TRAFFIC ON THE ROAD?
You ask: What is the comfort zone for the use of autonomous vehicles, and how can we expand this envelope over time? Perhaps we start with cars that only go 5 miles per hour and keep increasing the options as the technology advances. It’s also worth keeping in mind that humans driving and machines driving can be very different things. Autonomous vehicles with no passengers on board can perhaps steer in a jerky fashion or stop quickly and no one on board will mind. That way, we can lower the requirements and expand the options for what tasks to use self-driving cars, for example for driverless rebalancing trips.
DO YOU SEE A KILLER APP THAT WILL MAKE AUTONOMOUS DRIVING GO MAINSTREAM?
It’s not the “I want to read the paper while going work” idea. The killer app to my mind is reliable carsharing that changes the way we think of personal mobility. The knowledge that you’ll have transportation tomorrow at a certain time at a certain spot, and you can truly count on it. Then we can move from an ownership- to subscription-based model of cars. Most cars today are designed for long distances and high-speed driving. In fact, we use them mostly for city-driving. Personally, I hardly ever go further than five miles and rarely go faster than 5 mph. Yet at the same time they are not there yet to do many tasks without their drivers.
WILL THE ROAD TO AUTONOMOUS DRIVING BE ONE OF INCREMENTAL IMPROVEMENTS, OR ARE WE WAITING FOR SOME MAJOR BREAKTHROUGHS?
There are many pieces to this puzzle and they are all being developed at the same time. Just like rain drops form around dust particles in the sky – what scientists call nucleation – so it will be with advances in autonomous driving. There are little drops of technological progress here and there, like traffic jam or parking assistants, and as they become more numerous, they get bigger and eventually they come together and it rains. That’s a different notion than a giant leap forward, but this way, we don’t have to wait too long for autonomous driving to become a reality.