Smart Driving Cars episode 413- SpaceX, Waymo, NVIDIA, May Mobility & more
Smart Driving Cars Podcast
Smart Driving Cars episode 413- SpaceX, Waymo, NVIDIA, May Mobility & more
Annotations will appear as you listen
0:00
45:16
Car
Morgan Six Plus
The Morgan Plus Six is a small, two-seat sports car made by Morgan. It’s designed for driving for fun, with a focus on performance and a classic-looking style. People mention it when talking about how fast or exciting different sports cars are.
Term
Robo taxis
Robo taxis are self-driving cars that you can summon like a rideshare. They still rely on sensors to “see” the road, and bad weather or flooding can make that much harder.
Term
flooded roads
Flooded roads are dangerous for self-driving cars because the water can hide what’s under or on the road. It can also make the car lose grip and get stuck.
Term
heavy rain
Heavy rain makes it harder for self-driving cars to “see” clearly. It can also change what the road looks like and how slippery it is.
Company
Waymo
Waymo is a company that builds self-driving cars. Here, they’re mentioned because their service was paused after problems in bad weather and flooding.
Term
memory management
Memory management is how a computer program keeps track of its “working space” while it’s running. If it doesn’t manage that space well, the program can run out and stop doing the job it needs to do.
Concept
DARPA challenge
The DARPA Challenge was a competition that tested self-driving technology. The speaker is using it to explain how early software problems could stop a vehicle from completing a route.
Concept
world model (saving historic locations of detected objects)
A world model is the car’s internal “map” of what’s happening around it. Keeping track of where objects were seen helps it make better decisions, like whether it can back up safely.
Term
one camera looking forward
“One camera looking forward” means the car is mostly relying on a single front-facing view to understand the world. If you need to maneuver, you may not know what’s behind you as well.
Term
detecting objects
Object detection is how the car figures out what other things are around it. It’s like the car trying to label “there’s a car here” or “there’s an obstacle there.”
Company
Nvidia
Nvidia makes powerful computer chips. For self-driving cars, those chips help the car’s software “think” by running the AI that sees the road and plans what to do.
Company
May Mobility
May Mobility is a company that runs self-driving taxi/shuttle services. The discussion is about them getting more vehicles to expand where they operate.
Concept
driverless cars
Driverless cars are cars that can drive themselves without a person behind the wheel. The hosts are talking about laws that determine whether and where those cars are allowed to operate.
Concept
federal regulation
Federal regulation means national rules. For self-driving cars, the idea is that one set of rules could make it easier to deploy them consistently across many states.
LIVE
Welcome back to the Smart Driving Cars podcast. Thanks for spending time with us once again.
I'm Fred Fishkin along with the faculty chair of autonomous vehicle engineering at Princeton
University. Alan Kornhouser, hi Alan. Hey, good afternoon Fred. Good afternoon and we want to
remind everyone here at the top once again that you can find us on the transportation channel
on Fire TV, Roku and Apple TV these days. See us on the big screen. What an experience right
Alan? Oh my goodness. It's a great effort that we're happy to be part of. On top in the latest
newsletter Alan from Forbes, Harvard faculty votes to cap how many A's professors hand out.
It's being called an attempt to fight great inflation. I mean we've had great inflation
at Princeton for years and in fact a former dean of the college, she did all she could to try to
curtail that great inflation and it's only gotten worse. And I guess I'm glad to see
Harvard go out there and actually do something. This is the first thing Harvard's ever done.
I guess finally nice. Thank you, Harvard. But it's almost gotten silly.
We do understand that in fact our students never got a grade below an A before they came to Princeton.
Okay. I mean the competition for admission is really high and I'm sure it's the same thing at
Harvard. But you know it's suggested that by my colleague Bob Vanderby, I guess he really thought
of it first but I jumped on as soon as I heard it. You know instead of having A, B, C, D, F as
the grading system. Although the grading system used to be one through seven and one was the highest
grade. Seven was called flagrant neglect. In other words to get a seven. I mean if you didn't show
up you got a six. I mean you know birth to earn a seven. I mean you had to actually work on doing
nothing. Not show up and send you a postcard from the beach. Yeah that might be a six plus.
That might still not get you a seven. But Bob and I promoted maybe it should be A plus,
A minus, D, F. I mean it's still you know five levels and then everybody could be happy that you
know unless you get a deer or earned a deer and an F she's got an A minus so people on Wall
Street will be ooh must be good. Never mind. I don't know it's the emphasis put on grades by the
students is just it's just out of control and instead of you know being proud of what they know
or what they can figure out or what creativity they have or what you need uniqueness they have
it's this you know simple letter grade whatever which in the end has almost has very little
meaning compared to all the other things attributes that a student has.
So I don't know. Happy to see Harvard do that. Well there's another story Alan that I think
really hits close to home for you. Princeton now requiring instructors to be in exam rooms
for the first time since 1893 and I guess AI is getting much of the of the blame here.
Nearly a third of Princeton seniors admitted to cheating. I don't know who did the survey,
how it was done but you've got a link to a video from the House of Vell on this.
Yeah I mean
I mean I've been asking the question that one just hit me just knocked the wind out of me.
What are we doing admitting students to Princeton who cheat I mean whatever I mean
you know going through lifeline cheating and stealing this I mean that's I guess an easy way
to get money that really I mean you can live with yourself. I need to watch over you to make you
make sure you don't cheat what really that's what that's what's part of my job that's what
the university supposedly paying me to do. No university you don't have enough money
to pay me to do that. You just don't have enough money. I refuse to do it.
Don't take my class. Don't let me be your thesis advisor.
Don't ask me for advice. Plenty of people, other people. I'm not going to do it.
You're going to lie to me. You're going to cheat. Are you kidding?
What the heck are we here for?
I'll throw admissions under the bus.
What are we doing admitting kids that are going to cheat?
You mean all the AI and all the claw debts and all the chat-chip-eat poops and all
toots and whatever they are and so on can't help admissions until
did the applicant cheat in his or her application?
If so, they're off the list. We only admit you know this many out of that many.
Can we choose those you know throw out all those that cheat?
Can we figure that out? Likelihood of cheating?
And let's pick top of that group. What are we doing here?
And blaming it on AI. Blaming It's just not, no.
AI is a tool. Just like a steam shovel is a tool. Just like a steam locomotive is a tool.
Just like you know this thing is a tool. For you to you know take stuff that
each. It's not stuff that a human really should do.
Doesn't need to do.
If it does your long division for you, great.
You've taken some tests on you know second grade, you know, hey you know I do long division.
Now if you use you know the compute function on here, do long division,
we won't ask you to cite that. Okay.
No, our exams aren't. Hey, how well do you do division?
We don't care if you use long division or a calculator.
Because we have a deeper question to ask you on an examination.
And what's the purpose of an examination?
It's not really us. Not really for me to give you a grade.
It's for you to study the material again and learn it better. That's the reason.
And if you do that, everything will be fine.
If you don't use the examination to look at the material again and try to get a little bit
more out of it and try to learn something. Go to Harvard.
On the admissions question, Alan, I mean I know they they have for years
tried to put an emphasis on character, looking at activities and and accomplishments.
But I guess did you cheat on your activities?
Do you know how easy it is to cheat on somebody's activities? Who really did that?
Was it your parent?
Not that parents ever did stuff for their kids. Of course, I never did. They keep telling me.
No, that's not true either. But I this is just it's so disappointing.
Right. I'd like I'd like to I would like to see the question asked of those kids that
said that they they cheated at Princeton. Did you cheat to get into Princeton?
If you did, okay, you pulled one over on us. I think the administration, the whole faculty,
everybody here should be dead set on making sure nobody else comes to this university
who cheated to get in.
Let's say let's let's hear that from the from all as far as really I don't know what's
going to happen in the fall when I just tell them whoever's in the class in the fall. Hey,
if you're going to cheat, take something else. How many courses are being offered at Princeton
in the fall? Mine's not required.
And that will be your that's it. I'm intending that to be my opening statement.
Is that going to be year 55 now and how many years?
I've been there. I think my second year at Princeton. I don't know anything that's going on here.
I mean, it's so depressing.
I mean, I hope I hope that those that said admitted that they cheated at Princeton say
themselves, oh my goodness, what have I done to myself?
The proctoring wasn't about getting other kids to turn you in.
It was so you could look at yourself because guess what? You know you cheated.
You know you got what you got because you cheated. So guess what? You have to live for yourself.
You have to live with yourself.
Have a ball.
Right. Yipes. I'm just I'm distraught over it, but you know, whatever, you know,
works in me. I'm done, whatever.
Well, moving on in the newsletter, Alan, SpaceX has filed form S1 with the SEC.
Boys to the public. It's expected to be, I guess, the largest IPO in history.
Oh, who knows what I guess. I hope they didn't cheat in those that estimated whatever.
And I guess yesterday when they're scrub, I'm saying to myself, oh my goodness, please,
you know, don't push the button if you're not really sure because I could just see them just
screwing the whole thing up by blowing that thing up on the Fed and saying, oh my goodness,
we got to do it now. I just I thought a, I mean, that the president, that is a good organization.
As far as I look at, I look at their accomplishments. I look at what they have in
there. I look at what they're doing. I look at the way they're doing. I look at the people that
they put up there to basically tell us about what's going on. It is all, it is all as far as I'm
considered first class. And I can't believe they have cheaters in there. I just don't believe it.
Oh, but it's too long. Well, you're jealous. I get out of here. I mean, get whatever.
I don't know. I'm just, I'm just, I just hope that going public doesn't ruin them. But, you know,
so far, they just, if I look at what an entity has done, sure, okay, yes, it's, you know,
whatever, nothing is perfect, whatever. But if I, if I look at that, what that company's
done in terms of achievements, the good versus the nuts, I mean, you know,
they're, they're, it's a no brainer. And good luck. I'm happy to see that
people will make, will make some, some money off it. And if Elon makes money off it, God bless
them. I mean, gee, otherwise, we probably wouldn't have it at all. And I guess the point is Twitter
Twitter was a mess. Sorry, whatever. You know, is it really cleaned up? Is anything on the
inner tubes cleaned up? I don't know. I mean, look at what the heck's going on with some. I mean,
people are using this stuff to cheat. Come on, it's, it's a tool to help you, help you be more
creative, help you do the next job, have some shoulders to, to stand on, cite it, just say,
hey, it helped me. And here's what I created on top of it. That's what's good. Not here it is.
It's my work. That, that, that wasn't appropriate. That wasn't appropriate when we did it in libraries
and open up and took quotes out of books or the encyclopedia Britannica or any of these things.
It was inappropriate that it was inappropriate 500 years ago. It's always been inappropriate.
All right. Anyway, sorry, folks.
I think they can get a sense of Waymo was forced to pause service in four cities this week because
Robo taxis struggled to deal with heavy rain and flooded roads. There was a vehicle in Atlanta
that got stuck Wednesday while driving through a flooded street. And it just,
when we did the DARPA challenge in 2005,
unfortunately, we didn't have one line of code in our stuff. And therefore, we didn't do many
memory management properly. And after 9.8 miles, we ran out of memory. Why? Because we were saving
the location, historic location of every object we had detected to get that far. And why were we
saving the location of every object that we detected we went that far was because we could only afford
one camera looking forward. In case we had to go backwards, we say, what the hell is behind us?
Well, hey, if we save all the objects, we know where we are. We might actually be able to go in
reverse and not hit something and then find it different way out. I mean, that's the right reason
why we did it. Change one line of code, whatever. We went back out there three weeks later and we
said, oh, my goodness, with one line of code change, we're going to see if we can run the course.
Because we had the course. It was out there. We didn't need officials. Why not? We went out there
and we sort around the course.
But we didn't. We cheated a little. I actually took over. Why? Because in one of the
lake beds that the course went through,
it had rained in the intervening weeks. And now the course was actually flooded ahead.
And luckily, I was there to take over before it just went into the water.
And of course, I took over and I just went around it. Okay.
And since then, I've always thought of what kind of sensor would I need
to be able to determine the depth of water on the road ahead?
I don't think there exists one.
There you can put on a mobile device that's going to give you that information
in a finite amount of time so you could do anything about it.
I took over and went around that. Why? Because I did not know how deep it was.
I didn't know if it was this deep or that deep.
But I didn't know it was water.
And because I did not know I went around, every one of these things
needs to have that somewhere in its compute whatever.
Because you don't know. There isn't whatever you can't.
And basically, what tells me in terms of what we do handy rise in terms of our operational policy,
we're not going to run in thunderstorms of heavy rains.
With warnings about flooding.
Once high definition maps, they'd better know whether or not this stretch of road
is one that floods under those circumstances. I'm shocked that Wayne didn't know.
They knew the weather forecast. We do weather forecasts. They've really well.
Must have known. Were the people in the Philippines sleeping? All these things have to have oversight.
They don't provide oversight over their vehicles and say, hey, yo, it's under so I head up. Hey, cool it.
We're going to suspend service. New Jersey transit, suspend service on its rail lines all the time.
For whatever reason.
They don't promise the public that they're going to operate 24 7365.25.
What do we think these these things that can, you know, it really do work, but whatever.
Just in the best operational, the easiest operational design domains under the easiest
condition. How could how could Waymo have its vehicles out there actually moving?
That go into water. Why? Why was that really they were being attacked by water on both sides?
The whole thing was about the whatever. I mean,
come on. I mean, this is this is and while we're also not solving the heavy fog problem.
We shouldn't also be solving the curvious road problem.
We should not be solving the heavy snow problem. The governor of New Jersey all,
you know, any heavy snow around stay home. Yes, stay home.
These things aren't ready for be the mobility system of last resort.
It shouldn't be expected to be that we shouldn't have to wait until that they can do that before
we start getting some of the value out of this darn thing that so many people have spent so much
time and so much effort to create. Can't we just do a good job with some easy things and do that well
and be happy about that? Be happy that we're giving rides to people who really need a ride.
As opposed to some who want to say, oh my goodness, I'm going to be able to put it on the wood that
I had this thing in and because I insisted that it take me it went and got stuck in water. Are you
kidding? Since when is that an objective of all this stuff?
So in a sense, I feel sorry for poor Waymo. How they got themselves stuck and now they're not
pulled over when when weather conditions really suggested you shouldn't be doing that.
And you went into the water. There's an overlay on your code that doesn't take that much compute to
be able to. I don't know. I'm sorry. I'm sorry. And then, of course, the press has said, well,
my goodness, they can't go through a flooded road. I mean, how many cars in Atlanta were also
stuck in flooded roads for an hour that I mean, I guess they don't bother reporting, never mind.
Right. Boy, I'm in a mood today, aren't I? Let's see what we can. I think we're going to get you
in a better mood with the next one. I hope so. I'm depressed. So Val Oz recently confirmed as
assistant secretary of transportation for research and technology at the US DOT. She's slated to
deliver a keynote at an autonomous vehicle conference at the end of next week in Jacksonville.
Her keynote is titled. titled from pilot programs to scalable real world
deployment, building the affordable autonomy economy. I know you're looking forward to that.
Yeah. I mean, absolutely. We're barely across the start line of this thing.
And let's start getting some real value out of it because it has to scale or you don't get any
appreciable value if you don't scale. Okay. Because any of the things you're doing any value to or
just, you know, you just get a little bit, you know, just because somebody now has the opportunity
to go get their hair done and get back home, you know, that is very valuable to that person
that really improves their quality of life.
But in terms of the whole thing, it's a small amount. You got to do a lot of it.
There are a billion person trips today in the USA, a billion of them.
Okay. There are a billion of them every day. You want to improve on each of those, which is what
we're trying to do, let alone the 100 million more that are late in demand that stay home because,
you know, they're afraid that they won't be able to get back home. Probably is the main
reason why they stayed home. Why? Because they couldn't afford to do that. I don't have the right
whatever to do that. But the value is captured one at a time for those. So you have to scale.
So Paul is going to talk about that. She sees it. It's great that, you know, the person finally,
at least the whole thing will be legitimate that the head of the person that's the chief scientist,
essentially of DOT, got confirmed by the Senate. Diana, I don't think she was ever
confirmed. Robert, I don't think he was ever confirmed in previous two administrations.
So it's great. You know, US DOT is really ending up being, you know, the executive
department of the administration that's out there really trying to do the most good for
society in the US. I don't know at least, of course, that's my perspective. My perspective
is transportation and, you know, certainly it rings well in my ear. Things are happening.
Jensen Wong says Nvidia is powering Uber's push into a trillion-dollar self-driving
opportunity across 30 cities. Oh my goodness. I guess, you know, somebody's going to build
the vehicles for them and integrate all that stuff and get all that stuff working, and they're
going to come off the assembly line and we're going to do it. I'm wonderful. I'm thrilled.
I'm thrilled. And, you know, if it's Nvidia's providing the chips to do that,
they're probably the most. I, of course, think they're the most competent folks out there doing it.
I've loved watching them evolve from a gaming company to, I mean, my goodness,
what an evolution. Okay. But that's fantastic. Great. Can't wait to see it. Let's see. Why only
30 cities? How many cities are there in the United States? How many? What made those 30 cities the
important ones? We have more state capitals than that number. You mean every state capital doesn't
deserve it, plus every other? What? It's only the Manhattan's of the world. Are you kidding me?
Last place that you need these? Never mind. You don't want my view on that one. I mean,
they have a great subway system. Moving on, Ellen. I'm just trying to get the MTA to love me or
something like that. I don't think anybody's ever said, no, they do. If you have to move that many
people from where they are to where they want to go. Whoa. Right. Gonna have a heck of a mess if you
try to do it in the cars. It's not even conceivable. Maybe for the few that are so great. Let's give
one the transportation channel. Ellen, you've got a link in the newsletter to the latest
amusing zone mobility from our friend, Michael Senna, which we encourage everyone to read.
Absolutely. Just enjoy life. Read that. Michael L. Senna.com is the site. And from Bloomberg,
eCar X backed by Gillies Lee Shifu will sell Robo Taxis to May Mobility
in a $750 million deal that's going to involve, I think, thousands of vehicles.
Well, I hope so. 750 M divided by a thousand should be a few. There's a lot of money that
have available to buy those, at least some to put them out there to them. These things then are
productive elements in society, which actually should generate a return for May Mobility.
We've been watching it for some time. Great. Wonderful. Is Jensen going to power those too?
Why didn't he include May Mobility in his? Never mind.
A bill that would have allowed driverless cars in Minnesota did not pass as the legislature
there wrapped up the latest session. I guess disappointing to some, but states sometimes
fall behind. Yeah. Well, that's why we need the federal regulation, which is coming out. And
hopefully the federal regulations because these vehicles will need to be in interstate
commerce. I mean, I look at these things as commercial vehicles and the commercial vehicles
are either removing people or goods. And the commercial operation is the opportunity, whether
or not we end up owning them for our own whatever. Well, if we're going to own them,
they don't, because we, I mean, we will have to be there overseeing them and be the responsible
entity on them. Otherwise, I'm not sure that the public is ready to have their vehicles
be responsible enough to send their vehicles out there without them being responsible.
And if they're not there, I don't know. I just don't know how that plays out. So vehicles out
there that don't have the individual owners responsible have to be by a responsible corporate
entity that has, you know, somewhere with all to deal with the tail of the district. There's always
a tail of the distribution that has to be dealt with. Everything has a tail. I don't think the
individual is prepared to deal with the tail. Maybe they can buy services that deal with the tail,
but services that protect, deal with the tail, or those are expensive. So I don't know.
But a corporate entity whose mission is to be in that business, whose expertise is to be in
business, whose reputation is to be in that business, they might have a chance.
That's what Waymo and Uber and others are, I guess, you know, positioning themselves to do.
Maybe handy rods.
You alluded to the legislation in Washington, the surface
transportation reauthorization bill. You've got a link in the newsletter to the bill,
and I guess work is continuing. Yeah, work's continuing. So that's the, you know, first
form of that bill, what actually appears in the House and what ends up appearing in the Senate,
and then what passes, you know, as a compromise between the Senate and the House bills and gets
then, and gets signed by the president, we'll wait and see. But that's the opportunity to have
a unified U.S. approach to all this. And it really needs to be a U.S. approach. It almost
can't be a local state or local approach. It can start as a state local approach,
but to expect any kind, but to expect scale out of this. So yeah, great operates in San Francisco,
Silicon Valley area. Great, it operates in, you know, in the nice areas of Los Angeles and whatever,
but it's a big diverse country and the value of this and all the simulations that I've done
and all the looks that I've done with my students, which we think we have a nationwide view
that really lets us properly assess both rural suburban city type situations,
the need for mobility
beyond walking or beyond bum and ride in rural areas is just as strong as it is in
cities, whatever they are. And the opportunity for this to provide
high quality, affordable mobility to people who otherwise would be walking or bum and ride
is just, I don't know, it's what drives Elizabeth and me.
Terrific, Alan. Let's keep the progress going. Let's hope anything.
Yeah, I think, look, it seems as if we have developed an algorithmic
approach with digital computers and digital compute that is able to
figure out what's ahead and what's around and basically produce two fundamental outputs,
a steering angle, throttle position, just really two numbers.
And to generate those things
so fast, 30 times a second, so well,
that where I sit now,
it without a doubt drives me more safely than I can drive myself.
No, you know, maybe all the Alzheimer's and all the whatever physical
one to do is like gone. It's not even close
in situations where, you know,
it's not a hurricane sandy. It's not a torrential downpour.
It's not fog that, you know, you can't even see your hand.
It's not snow that can't even plow through it.
It's not that I want to drive down the stream or ravine or climb the great wall of China with my vehicle.
We even care to drive or be taken down the curviest road in the world in San Francisco.
Although that thing happens to be pretty darn easy.
It's a shame that we really are not completely gung-ho about making this available to people who
there are lots of people today who either don't want to drive
aren't allowed to drive, can't drive,
can't afford to drive, drive is probably the biggest one of those.
It can now be given a ride,
dignified ride, not be begging for it, not have to alter their life. Oh, I can only do it then.
How about it be demand responsive as opposed to supply responsive?
In all the mobility systems we have out there that drive our own cars are all
they're all supply responsive. It's what the supplier, it's the benefit the person putting
the supply out there, not the benefit the consumer. Is there been any economy in which
the supply responsive solution was to beat out the demand responsive solution?
I don't know, I can't think of one, but I'm sure those of you out there can maybe.
That's a shame that, oh my goodness, I've been given a ride to a lot of people lately.
There hasn't been one of them that said, oh my goodness,
who, this is pretty damn interesting. Anyway, but maybe I've by some, it's by sample. Hey,
I didn't, ooh, I didn't put it in a peer review, whatever. I don't know.
Since Elizabeth and I are just doing it for ourselves, I guess, you know, we're just trying
to see, we're just trying to learn ourselves. We're just trying to educate ourselves.
But man, who looks pretty darn good anyway. You and lots of others are learning firsthand, right?
Well, we're trying. I mean, we've tried. I've tried for, I've failed for all these years. It's about
time that I failed some more. Why not? I'm going to end on a high note, Alan, recommendation.
Got a book recommendation, The Laws of Thought, The Quest for a Mathematical Theory of the Mind
by Tom Griffiths, a professor of psychology and computer science there at Princeton.
Yep. He teaches a cognitive science course. It's really a great course. I think he's put
together a great book, that and a Nails book that I've sort of put out there. If you're involved in
any of this stuff, you really should read. Those are two nice reads and you should take a look at them.
Absolutely. And we're in reunions here at Princeton. Princeton reunions are, you know,
unique. It's nice to see a lot of former graduates and so on. It's a good time and
men's lacrosse is in its final four with Duke tomorrow. Got a couple of students on the team,
so wish them the best. Oh, life's good. Very, very excited. We wish them good luck.
Yeah. Unfortunately, the finals, if they win is on class day. And I can't, I'm
You can find more and the newsletter at smartdrivingcar.com. And we told you,
you can find us at the Transportation Channel. You can find my tech reports at
textination.com. Thank you again for taking the time to watch or listen. Have a great holiday
weekend and stay safe. Happy holiday. Happy coming of summer.
About this episode
The conversation jumps from real-world autonomy limits to the policy and compute behind scaling. Waymo is said to have paused service in four cities after heavy rain and flooded roads, prompting debate about whether systems can rely on HD maps and handle deep water. They connect these reliability gaps to regulation—state bills like Minnesota’s failing, and hopes for federal rules for interstate commerce. The hosts also cover Nvidia powering Uber’s expansion, May Mobility’s robotaxi deal, and how algorithmic control outputs can be generated fast for safer driving.
Original notes
SpaceX files for an IPO, Waymo pauses for flooding, Seval Oz to keynote an AV Conference plus NVIDIA, Uber, May Mobility and more. Join Princeton's Alain Kornhauser and co-host Fred Fishkin for episode 413 of Smart Driving Cars!
Smart Driving Cars Podcast
Fred Fishkin/Alain Kornhauser
0:00
45:16