WEBVTT 00:00:00.703 --> 00:00:02.617 Let's face it: 00:00:02.617 --> 00:00:05.062 Driving is dangerous. 00:00:05.062 --> 00:00:08.160 It's one of the things that we don't like to think about, 00:00:08.160 --> 00:00:11.812 but the fact that religious icons and good luck charms 00:00:11.812 --> 00:00:16.602 show up on dashboards around the world 00:00:16.602 --> 00:00:20.739 betrays the fact that we know this to be true. 00:00:20.739 --> 00:00:24.333 Car accidents are the leading cause of death 00:00:24.333 --> 00:00:28.503 in people ages 16 to 19 in the United States -- 00:00:28.503 --> 00:00:31.346 leading cause of death -- 00:00:31.346 --> 00:00:35.209 and 75 percent of these accidents have nothing to do 00:00:35.209 --> 00:00:37.494 with drugs or alcohol. NOTE Paragraph 00:00:37.494 --> 00:00:39.755 So what happens? 00:00:39.755 --> 00:00:43.974 No one can say for sure, but I remember my first accident. 00:00:43.974 --> 00:00:47.777 I was a young driver out on the highway, 00:00:47.777 --> 00:00:50.035 and the car in front of me, I saw the brake lights go on. 00:00:50.035 --> 00:00:51.835 I'm like, "Okay, all right, this guy is slowing down, 00:00:51.835 --> 00:00:53.117 I'll slow down too." 00:00:53.117 --> 00:00:55.043 I step on the brake. 00:00:55.043 --> 00:00:57.297 But no, this guy isn't slowing down. 00:00:57.297 --> 00:01:00.475 This guy is stopping, dead stop, dead stop on the highway. 00:01:00.475 --> 00:01:03.015 It was just going 65 -- to zero? 00:01:03.015 --> 00:01:04.535 I slammed on the brakes. 00:01:04.535 --> 00:01:07.594 I felt the ABS kick in, and the car is still going, 00:01:07.594 --> 00:01:10.290 and it's not going to stop, and I know it's not going to stop, 00:01:10.290 --> 00:01:13.229 and the air bag deploys, the car is totaled, 00:01:13.229 --> 00:01:16.786 and fortunately, no one was hurt. 00:01:16.786 --> 00:01:20.997 But I had no idea that car was stopping, 00:01:20.997 --> 00:01:24.642 and I think we can do a lot better than that. 00:01:24.642 --> 00:01:28.787 I think we can transform the driving experience 00:01:28.787 --> 00:01:32.666 by letting our cars talk to each other. NOTE Paragraph 00:01:32.666 --> 00:01:34.090 I just want you to think a little bit 00:01:34.090 --> 00:01:36.978 about what the experience of driving is like now. 00:01:36.978 --> 00:01:41.006 Get into your car. Close the door. You're in a glass bubble. 00:01:41.006 --> 00:01:43.922 You can't really directly sense the world around you. 00:01:43.922 --> 00:01:46.103 You're in this extended body. 00:01:46.103 --> 00:01:48.266 You're tasked with navigating it down 00:01:48.266 --> 00:01:50.322 partially-seen roadways, 00:01:50.322 --> 00:01:54.746 in and amongst other metal giants, at super-human speeds. 00:01:54.746 --> 00:01:59.226 Okay? And all you have to guide you are your two eyes. 00:01:59.226 --> 00:02:00.988 Okay, so that's all you have, 00:02:00.988 --> 00:02:02.723 eyes that weren't really designed for this task, 00:02:02.723 --> 00:02:06.474 but then people ask you to do things like, 00:02:06.474 --> 00:02:08.023 you want to make a lane change, 00:02:08.023 --> 00:02:10.344 what's the first thing they ask you do? 00:02:10.344 --> 00:02:13.439 Take your eyes off the road. That's right. 00:02:13.439 --> 00:02:15.535 Stop looking where you're going, turn, 00:02:15.535 --> 00:02:17.553 check your blind spot, 00:02:17.553 --> 00:02:21.024 and drive down the road without looking where you're going. 00:02:21.024 --> 00:02:24.159 You and everyone else. This is the safe way to drive. 00:02:24.159 --> 00:02:26.400 Why do we do this? Because we have to, 00:02:26.400 --> 00:02:28.979 we have to make a choice, do I look here or do I look here? 00:02:28.979 --> 00:02:30.500 What's more important? 00:02:30.500 --> 00:02:33.211 And usually we do a fantastic job 00:02:33.211 --> 00:02:36.980 picking and choosing what we attend to on the road. 00:02:36.980 --> 00:02:40.630 But occasionally we miss something. 00:02:40.630 --> 00:02:45.091 Occasionally we sense something wrong or too late. 00:02:45.091 --> 00:02:47.079 In countless accidents, the driver says, 00:02:47.079 --> 00:02:49.387 "I didn't see it coming." 00:02:49.387 --> 00:02:52.668 And I believe that. I believe that. 00:02:52.668 --> 00:02:55.593 We can only watch so much. NOTE Paragraph 00:02:55.593 --> 00:03:00.737 But the technology exists now that can help us improve that. 00:03:00.737 --> 00:03:05.033 In the future, with cars exchanging data with each other, 00:03:05.033 --> 00:03:08.961 we will be able to see not just three cars ahead 00:03:08.961 --> 00:03:10.555 and three cars behind, to the right and left, 00:03:10.555 --> 00:03:13.721 all at the same time, bird's eye view, 00:03:13.721 --> 00:03:16.849 we will actually be able to see into those cars. 00:03:16.849 --> 00:03:19.220 We will be able to see the velocity of the car in front of us, 00:03:19.220 --> 00:03:22.460 to see how fast that guy's going or stopping. 00:03:22.460 --> 00:03:26.970 If that guy's going down to zero, I'll know. NOTE Paragraph 00:03:26.970 --> 00:03:30.829 And with computation and algorithms and predictive models, 00:03:30.829 --> 00:03:34.102 we will be able to see the future. 00:03:34.102 --> 00:03:35.658 You may think that's impossible. 00:03:35.658 --> 00:03:38.389 How can you predict the future? That's really hard. 00:03:38.389 --> 00:03:42.008 Actually, no. With cars, it's not impossible. 00:03:42.008 --> 00:03:44.740 Cars are three-dimensional objects 00:03:44.740 --> 00:03:47.072 that have a fixed position and velocity. 00:03:47.072 --> 00:03:48.703 They travel down roads. 00:03:48.703 --> 00:03:51.115 Often they travel on pre-published routes. 00:03:51.115 --> 00:03:55.053 It's really not that hard to make reasonable predictions 00:03:55.053 --> 00:03:57.917 about where a car's going to be in the near future. 00:03:57.917 --> 00:03:59.919 Even if, when you're in your car 00:03:59.919 --> 00:04:01.913 and some motorcyclist comes -- bshoom! -- 00:04:01.913 --> 00:04:04.209 85 miles an hour down, lane-splitting -- 00:04:04.209 --> 00:04:06.756 I know you've had this experience -- 00:04:06.756 --> 00:04:09.359 that guy didn't "just come out of nowhere." 00:04:09.359 --> 00:04:13.002 That guy's been on the road probably for the last half hour. 00:04:13.002 --> 00:04:14.192 (Laughter) 00:04:14.192 --> 00:04:17.781 Right? I mean, somebody's seen him. 00:04:17.781 --> 00:04:20.549 Ten, 20, 30 miles back, someone's seen that guy, 00:04:20.549 --> 00:04:22.933 and as soon as one car sees that guy 00:04:22.933 --> 00:04:25.164 and puts him on the map, he's on the map -- 00:04:25.164 --> 00:04:27.340 position, velocity, 00:04:27.340 --> 00:04:29.661 good estimate he'll continue going 85 miles an hour. 00:04:29.661 --> 00:04:31.845 You'll know, because your car will know, because 00:04:31.845 --> 00:04:34.120 that other car will have whispered something in his ear, 00:04:34.120 --> 00:04:36.043 like, "By the way, five minutes, 00:04:36.043 --> 00:04:38.818 motorcyclist, watch out." 00:04:38.818 --> 00:04:41.521 You can make reasonable predictions about how cars behave. 00:04:41.521 --> 00:04:42.886 I mean, they're Newtonian objects. 00:04:42.886 --> 00:04:45.795 That's very nice about them. NOTE Paragraph 00:04:45.795 --> 00:04:48.829 So how do we get there? 00:04:48.829 --> 00:04:51.095 We can start with something as simple 00:04:51.095 --> 00:04:53.965 as sharing our position data between cars, 00:04:53.965 --> 00:04:55.857 just sharing GPS. 00:04:55.857 --> 00:04:58.301 If I have a GPS and a camera in my car, 00:04:58.301 --> 00:05:00.532 I have a pretty precise idea of where I am 00:05:00.532 --> 00:05:02.264 and how fast I'm going. 00:05:02.264 --> 00:05:03.921 With computer vision, I can estimate where 00:05:03.921 --> 00:05:07.458 the cars around me are, sort of, and where they're going. 00:05:07.458 --> 00:05:08.428 And same with the other cars. 00:05:08.428 --> 00:05:10.242 They can have a precise idea of where they are, 00:05:10.242 --> 00:05:12.388 and sort of a vague idea of where the other cars are. 00:05:12.388 --> 00:05:15.619 What happens if two cars share that data, 00:05:15.619 --> 00:05:17.574 if they talk to each other? 00:05:17.574 --> 00:05:20.352 I can tell you exactly what happens. 00:05:20.352 --> 00:05:22.691 Both models improve. 00:05:22.691 --> 00:05:24.746 Everybody wins. 00:05:24.746 --> 00:05:27.323 Professor Bob Wang and his team 00:05:27.323 --> 00:05:30.061 have done computer simulations of what happens 00:05:30.061 --> 00:05:33.492 when fuzzy estimates combine, even in light traffic, 00:05:33.492 --> 00:05:36.116 when cars just share GPS data, 00:05:36.116 --> 00:05:38.629 and we've moved this research out of the computer simulation 00:05:38.629 --> 00:05:41.656 and into robot test beds that have the actual sensors 00:05:41.656 --> 00:05:44.789 that are in cars now on these robots: 00:05:44.789 --> 00:05:46.627 stereo cameras, GPS, 00:05:46.627 --> 00:05:48.501 and the two-dimensional laser range finders 00:05:48.501 --> 00:05:50.741 that are common in backup systems. 00:05:50.741 --> 00:05:55.225 We also attach a discrete short-range communication radio, 00:05:55.225 --> 00:05:57.134 and the robots talk to each other. 00:05:57.134 --> 00:05:58.673 When these robots come at each other, 00:05:58.673 --> 00:06:01.644 they track each other's position precisely, 00:06:01.644 --> 00:06:04.381 and they can avoid each other. NOTE Paragraph 00:06:04.381 --> 00:06:07.607 We're now adding more and more robots into the mix, 00:06:07.607 --> 00:06:09.078 and we encountered some problems. 00:06:09.078 --> 00:06:11.437 One of the problems, when you get too much chatter, 00:06:11.437 --> 00:06:15.165 it's hard to process all the packets, so you have to prioritize, 00:06:15.165 --> 00:06:17.522 and that's where the predictive model helps you. 00:06:17.522 --> 00:06:21.894 If your robot cars are all tracking the predicted trajectories, 00:06:21.894 --> 00:06:23.661 you don't pay as much attention to those packets. 00:06:23.661 --> 00:06:25.364 You prioritize the one guy 00:06:25.364 --> 00:06:26.697 who seems to be going a little off course. 00:06:26.697 --> 00:06:29.223 That guy could be a problem. 00:06:29.223 --> 00:06:32.225 And you can predict the new trajectory. 00:06:32.225 --> 00:06:34.988 So you don't only know that he's going off course, you know how. 00:06:34.988 --> 00:06:38.713 And you know which drivers you need to alert to get out of the way. NOTE Paragraph 00:06:38.713 --> 00:06:41.346 And we wanted to do -- how can we best alert everyone? 00:06:41.346 --> 00:06:44.529 How can these cars whisper, "You need to get out of the way?" 00:06:44.529 --> 00:06:46.046 Well, it depends on two things: 00:06:46.046 --> 00:06:48.215 one, the ability of the car, 00:06:48.215 --> 00:06:51.432 and second the ability of the driver. 00:06:51.432 --> 00:06:52.937 If one guy has a really great car, 00:06:52.937 --> 00:06:55.862 but they're on their phone or, you know, doing something, 00:06:55.862 --> 00:06:57.792 they're not probably in the best position 00:06:57.792 --> 00:07:00.762 to react in an emergency. 00:07:00.762 --> 00:07:02.427 So we started a separate line of research 00:07:02.427 --> 00:07:04.978 doing driver state modeling. 00:07:04.978 --> 00:07:07.307 And now, using a series of three cameras, 00:07:07.307 --> 00:07:09.577 we can detect if a driver is looking forward, 00:07:09.577 --> 00:07:12.437 looking away, looking down, on the phone, 00:07:12.437 --> 00:07:15.498 or having a cup of coffee. 00:07:15.498 --> 00:07:17.568 We can predict the accident 00:07:17.568 --> 00:07:21.219 and we can predict who, which cars, 00:07:21.219 --> 00:07:24.705 are in the best position to move out of the way 00:07:24.705 --> 00:07:27.714 to calculate the safest route for everyone. 00:07:27.714 --> 00:07:32.349 Fundamentally, these technologies exist today. NOTE Paragraph 00:07:32.349 --> 00:07:35.173 I think the biggest problem that we face 00:07:35.173 --> 00:07:38.186 is our own willingness to share our data. 00:07:38.186 --> 00:07:40.817 I think it's a very disconcerting notion, 00:07:40.817 --> 00:07:43.203 this idea that our cars will be watching us, 00:07:43.203 --> 00:07:46.574 talking about us to other cars, 00:07:46.574 --> 00:07:50.001 that we'll be going down the road in a sea of gossip. 00:07:50.001 --> 00:07:53.898 But I believe it can be done in a way that protects our privacy, 00:07:53.898 --> 00:07:57.639 just like right now, when I look at your car from the outside, 00:07:57.639 --> 00:08:00.002 I don't really know about you. 00:08:00.002 --> 00:08:01.139 If I look at your license plate number, 00:08:01.139 --> 00:08:03.025 I don't really know who you are. 00:08:03.025 --> 00:08:07.274 I believe our cars can talk about us behind our backs. NOTE Paragraph 00:08:07.274 --> 00:08:10.249 (Laughter) NOTE Paragraph 00:08:10.249 --> 00:08:13.434 And I think it's going to be a great thing. 00:08:13.434 --> 00:08:15.084 I want you to consider for a moment 00:08:15.084 --> 00:08:19.202 if you really don't want the distracted teenager behind you 00:08:19.202 --> 00:08:21.322 to know that you're braking, 00:08:21.322 --> 00:08:24.246 that you're coming to a dead stop. 00:08:24.246 --> 00:08:26.987 By sharing our data willingly, 00:08:26.987 --> 00:08:29.799 we can do what's best for everyone. NOTE Paragraph 00:08:29.799 --> 00:08:32.875 So let your car gossip about you. 00:08:32.875 --> 00:08:35.913 It's going to make the roads a lot safer. NOTE Paragraph 00:08:35.913 --> 00:08:37.704 Thank you. NOTE Paragraph 00:08:37.704 --> 00:08:42.689 (Applause)