Self-Driving Cars Need Lessons On Human Drivers | Maddie About Science

[00:00:01]Maddie: There’s my speed limit.

[00:00:02]I’m just going to go ahead … Oh no, do I not?

[00:00:04]Hey!

[00:00:05](screeching car brakes) Driving can be pretty stressful.

[00:00:10]Especially when people out there can be really unpredictable.

[00:00:15]And soon we’ll be dealing with a new type of driver.

[00:00:18]Over the last several years, billions of dollars have been invested in developing self-driving cars.

[00:00:24]So, like it or not, most of us will be sharing the road with autonomous vehicles in the future.

[00:00:30]But how are those logical robots going to learn to share the road with human drivers that don’t always follow the rules?

[00:00:38]To find out, we headed to Stanford University, located in the heart of Silicon Valley.

[00:00:44]It's a hot spot for autonomous vehicle technology.

[00:00:47]There, I met up with Dorsa Sadigh.

[00:00:49]She’s a computer scientist studying how people drive.

[00:00:53]Her goal is to use that information to help self-driving cars better predict human behavior.

[00:00:59]Let’s just think of her as a human ambassador for robots.

[00:01:03]Maddie: I assume you’ve been in self-driving cars, right?

[00:01:05]Dorsa: Yeah, a couple times.

[00:01:07]Maddie: What’s it like?

[00:01:08]Dorsa: It’s kind of like sitting right next to someone else who you’re not comfortable with his driving or her driving.

[00:01:15]Like it reminds me of sitting right next to my husband.

[00:01:18]Maddie: I hope he’s watching.

[00:01:19]Get it together, Dorsa’s husband.

[00:01:21]Maddie: What do you think some of the challenges are going to be as self-driving cars start to come into our everyday lives?

[00:01:27]Dorsa: Safety is a big, big, big problem that we still have with autonomous cars.

[00:01:31]These cars are OK with driving in normal driving conditions and normal roads with normal obstacles around them where they can detect everything.

[00:01:39]The moment you put them in situations that they haven’t seen, they don’t really know how to deal with that.

[00:01:43]If all cars were autonomous, a lot of our problems would have been solved because it would have been much safer and easier for the cars to coordinate with each other, communicate with each other and know what they are doing and not necessarily deal with this other agent, this human driver.

[00:01:55]Maddie: Before Dorsa can teach robots what to expect from human drivers, she has to study how humans drive.

[00:02:01]So she took me to a pretty sweet driving simulator.

[00:02:04]We’re talking 360-degree screen, real car, and a lot of sensors, all dedicated to learning how people drive.

[00:02:12]Hey, how do I look?

[00:02:14]Good?

[00:02:16]All right, cool.

[00:02:17]Dorsa: Oh, seat belt?

[00:02:18]Maddie: Oh, yeah, sure Dorsa.

[00:02:19](laughter, seatbelt clicks) All right, you ready to cook?

[00:02:23]Oh, look at this speedster.

[00:02:25]What are we measuring of me right now?

[00:02:27]What are you getting?

[00:02:28]Dorsa: So the main thing we are getting is your steering angle and your acceleration braking, so that gives us your trajectory.

[00:02:34]Maddie: My acceleration and my braking?

[00:02:35]Dorsa: Yeah.

[00:02:36]Maddie: OK.

[00:02:36]And check out that orange dot.

[00:02:38]The goggles I’m wearing allow them to track the movement of my eyeballs during the simulation.

[00:02:43]It helps them determine what I’m paying attention to behind the wheel … right before they try to kill me.

[00:02:49]Hey!

[00:02:50](Car brakes screeching) Did you see that?

[00:02:52]Dorsa: Good job though.

[00:02:53]Maddie: That guy almost killed us.

[00:02:55]Good try, buddy.

[00:02:57]Dorsa: So the other thing that we do is actually we create near-accident scenarios to see how humans respond.

[00:03:02]Maddie: Well that felt like a near-accident scenario, Dorsa.

[00:03:04]Dorsa: You responded well, yeah you braked.

[00:03:06]Maddie: Well, thank you.

[00:03:07]Oh doggie!

[00:03:08]Doggie!

[00:03:09]Somebody microchip that dog.

[00:03:11]It doesn’t even have a collar on.

[00:03:12]All right see you motorcycle.

[00:03:14]Hey!

[00:03:15]Easy buddy.

[00:03:18]That motorcycle is trouble, did you see that?

[00:03:19]Hey hey hey!

[00:03:20]Simulator voice: Proceed straight at the intersection.

[00:03:22]Maddie: What are people doing around here?

[00:03:25]What’s kind of the endgame with your research?

[00:03:27]What are you hoping that your research group can contribute to society and autonomous vehicles?

[00:03:32]Dorsa: Yeah, so I’m really hoping by modeling humans and modeling this interaction between human-driven cars and autonomous cars we can have better and more efficient autonomous cars driving on shared roads with other humans around them.

[00:03:43]Maddie: Crazy humans like me.

[00:03:45]Dorsa’s long-term goal is to build a model of human driving behavior that can be programmed into self-driving cars.

[00:03:53]Go ahead.

[00:03:54]That’s just a couple of humans interacting.

[00:03:56]Enjoy your day, fellow human.

[00:03:59]Maddie: Hey everybody, thanks for watching.

[00:04:00]Adam: Subscribe to Skunk Bear, NPR’s science show.

[00:04:02]Maddie: And check out the last episode of Maddie About Science.

[00:04:04]Oh, it’s a red light.

[00:04:06]Maddie, no!

[00:04:07]Nailed it!