Jennifer Healey is a PhD in computer science from MIT . She works at Intel Corporation Research Labs and researches devices and systems that will enable major innovations. In a talk she gave for TED Talks titled “If cars could talk, accidents might be avoidable,” she presented how she imagines a world without (traffic) accidents.
The translation into Greek was done by Nikolao Benia and edited by Dimitri Katevati.
Watch the video
Let's face it: Driving is dangerous. It's one of those things we don't want to think about, but the fact that religious images and amulets are placed on dashboards around the world betrays the fact that we know this to be true. Car accidents are the leading cause of death for 16- to 19-year-olds in the United States — the leading cause of death — and 75 percent of those accidents are not drug or alcohol related.
So what happens? No one can say for sure, but I remember my first accident. I was a new driver on the highway and I saw the brake lights on the car in front of me come on. I think, "Okay, that's okay, the guy is slowing down, I'm going to slow down too." I hit the brakes. But no, the guy didn't slow down. The guy was stopping, completely, completely stopping on the highway. Did he go from 60 to 0? I hit the brakes really hard. I felt the ABS kick in and the car was still going, and it wasn't going to stop, and I know it wasn't going to stop, and the airbag deployed, the car was torn apart, and thankfully, no one was hurt. But I had no idea that car was going to stop, and I think we can do a lot better than that. I think we can transform the driving experience by letting our cars talk to each other.
I want you to think for a moment about what the driving experience is like now. You get in the car. You close the door. You're inside a glass bubble. You can't directly sense the world around you. You're inside an extension of yourself. You have to navigate it on partially visible highways, in and between other metal giants, at superhuman speeds. Right? And the only things guiding you are your two eyes. Right, those are the only things you have, eyes that really aren't designed for that purpose, but people ask you to do things, like when you want to change lanes, what's the first thing they ask you to do? Take your eyes off the road. Right. Stop looking where you're going, turn around, look in your blind spot, and keep driving without looking where you're going. You and everyone else. That's the safe way to drive. Why do we do this? Because we have to, we have to choose, do I see here or do I see there? Which is more important? And we usually do an amazing job of picking and choosing what to look for on the road. But, occasionally something will slip through our fingers. Occasionally we sense something either wrongly or too late. In countless accidents, the driver says, "I didn't see that coming." And I believe that. I believe that. That's how much we can pay attention.
But now there's technology that can help us improve that. In the future, with cars sharing data with each other, we'll be able to see not just three cars in front and three cars behind, to the right and left, all at the same time, panoramic views. We'll be able to see inside those cars. We'll be able to see the speed of the car in front, see how fast someone's going or if they're stopping. If someone's coming to a complete stop, I'll know.
And with calculations, algorithms, and predictive models, we'll be able to see into the future. You might think that's impossible. How do you predict the future? That's very difficult. Actually, no. For cars, it's not impossible. Cars are three-dimensional objects that have a specific position and speed. They move on a road. They often follow predetermined routes. It's really not that hard to make reasonable predictions about where a car is going to be in the near future. Even if you're in a car and a motorcyclist comes along -- boom! -- going 87 miles an hour, changing lanes -- I know you've had that experience -- this guy didn't "come out of nowhere." This guy has probably been on the road for the last half hour. (Laughter) Right? I mean, someone has seen him. 15, 30, 50 kilometers before, someone has seen it, and once a car sees it and places it on the map, it's on the map — location, speed, a good estimate that it's going to keep going at 140 kilometers per hour. You'll know it, because your car will know it, because the other car whispered something in its ear, like, "By the way, five minutes, motorcyclist, watch out." You can make reasonable predictions about how cars behave. I mean, they're Newtonian objects. That's the great thing about them.
So how do we get there? We can start with something as simple as sharing our location data between cars, just sharing GPS. If I have a GPS and a camera in my car, I have a pretty accurate idea of where I am and how fast I'm going. With machine vision, I can estimate where the cars around me are, roughly, and where they're going. Same with other cars. They can have a precise idea of where they are and a vague idea of where the other cars are. What happens when two cars share this data, if they talk to each other? I can tell you exactly what happens. Both models get better. Everyone wins. Professor Bob Wang and his team have done computer simulations of what happens when fuzzy estimates are combined, even in light congestion, when cars are simply sharing GPS data, and we've taken this research outside of computer simulations to robot testbeds that have the actual sensors that are now in cars, in these robots: stereo cameras, GPS, and two-dimensional laser rangefinders, which are common in backup systems. We also adapted a discreet short-range radio, and the robots talk to each other. When these robots get close to each other, they accurately track each other's location and can avoid each other.
Now we're adding even more robots to the space, and we've run into some problems. One of the problems, when you get a lot of chatter, it's hard to process all the packets, so you have to prioritize, and that's where the predictive model comes in. If the robot cars are following the predicted paths, you don't give much importance to those packets. You give priority to the one that seems to be going a little bit off course. That guy can create a problem. And you can predict a new path. So not only do you know that it's going to go off course, you know how. And you know which drivers you need to warn to move over.
And we wanted to do -- what's the best way to alert everyone? How can cars whisper, "You need to pull over?" It depends on two things: first, the capability of the car and second, the capability of the driver. If someone has a really good car, but they're using their phone or, you know, doing something, they're probably not in the best position to react to an emergency. So we started a separate study modeling the driver's state. Now, using a three-camera array, we can detect whether the driver is looking ahead, looking away, looking down, on their phone or drinking coffee. We can predict the accident, and we can predict who, which cars, are in the best position to move off-track and figure out the safest route for everyone. Basically, these technologies exist today.
I think the biggest problem we face is our willingness to share our data. I think it's a very disturbing concept, the idea that our cars will be watching us, talking about us to other cars, gossiping about us all the time. But I think it can be done in a way that protects our privacy, just like right now, when I look at the outside of your car, I really don't know about you. If I look at your license plate, I don't know who you are. I think our cars will be talking about us behind our backs.
(Laughter)
And I think it's going to be a really good thing. I want you to think for a moment about whether you really don't want the absent-minded teenager behind you to know that you're braking, that you're planning to stop. By voluntarily sharing our data, we can do what's best for everyone.
So let your car gossip for you. It will make the roads safer.
Thank you.
