- 1 hour ago
Waymo, Tesla and others are racing to build a potentially trillion-dollar robotaxi business. But after riding in both cars, visiting Waymo's San Francisco depot, interviewing safety engineers, and activists — we discovered that nothing on the road today is truly autonomous. In this episode of The Limit, we explore all the reasons why self-driving cars are almost impossible.
Category
😹
FunTranscript
00:00This is my first time inside of a Waymo and we're going down the windiest street in San
00:04Francisco.
00:05Even though it looks like no one's driving this car, not this Waymo, not Tesla in full
00:10self-driving, none of these cars are fully autonomous no matter how good they seem to
00:15be at navigating the roads.
00:17Waymo's only operate in specific geofence locations, and Tesla's full self-driving
00:24mode requires eyes on the road and hands by the wheel.
00:27Whoa, don't do that.
00:29And when any of these things get stuck, they require the intervention of a human to tell
00:34it what to do.
00:35Hello, this is Casey with Waymo Cleaver Pounds.
00:38So I should just close the door?
00:39Yeah.
00:40Okay, cool.
00:41I'm going to do that.
00:41Self-driving cars are coming soon to a city near you.
00:46But we still don't know how many humans it takes to drive them.
00:50So why is it so hard to get robo-taxis right?
00:53And which company could be the first to unlock this trillion-dollar market?
01:01There are two main players already vying to bring robo-taxis to American roads.
01:06Waymo and Tesla have two very different approaches to autonomy.
01:10And when you see them side by side, that's extremely clear.
01:14The Waymo has LiDAR, radar, 29 cameras all around the vehicle.
01:19The LiDAR spins out laser pulses to build a high-resolution map of nearby cars, cyclists,
01:25and pedestrians.
01:26Radar uses radio waves to track how fast objects are moving and how far away they are, even
01:30through rain, fog, and dust.
01:32Cameras read colors, signs, and signals.
01:35In case you're wondering what happens if you touch one of these things, literally nothing.
01:40It's spinning at a speed that's not unsafe, and it doesn't break it to do that.
01:45So, but don't do that at home.
01:46Tesla has just eight cameras, barely visible at all.
01:51Elon Musk has called LiDAR a fool's errand and something that is expensive and unnecessary.
01:56Tesla's camera-only approach has an advantage, cost.
02:00It's way cheaper to have a self-driving car navigate with cameras and AI than to have all
02:06the additional sensors that a Waymo does.
02:08But there's a downside.
02:10Some of the experts I spoke to said the camera-only approach will never truly be safe.
02:16If you're serious about self-driving, you don't have just a single sensor.
02:21Using camera-only is like tying one hand behind your back when you could also use a LiDAR.
02:25It's certainly taking the hard path.
02:27And why that matters is because this car right here, if Tesla can prove that this car is safe
02:33enough and can self-drive everywhere in the country, every Tesla on the road today could
02:39become a cybertaxi overnight. Fundamentally, two different approaches to how self-driving cars navigate.
02:46Please close the right side rear door. Thanks.
02:50Now let's talk about what makes them different from human drivers.
02:55There's a very long list of fairly normal driving scenarios that robo-taxis struggle with
03:01that aren't an issue for any normally skilled human driver. Here's one.
03:09Tesla's don't always stop for school buses.
03:14It just carries on.
03:16It was 100%. When we first started testing, it was 100% of the time. That's Dan O'Dowd. He's a
03:23critical
03:24systems engineer who designed the software that runs in F-35 fighter jets and nuclear bombers.
03:31The kids would line up and watch us run over the mannequins, the kids in the school. Their parents would
03:36come out or the teachers would come out and say, oh my god, it ran over that little kid who's
03:40just
03:40walking across the road. Over the years, Dan has probably spent millions of his own dollars trying
03:46to show how Tesla's full self-driving software has major design flaws. Like Tesla's driving on the wrong
03:54side of the road. There can be a do not enter sign on both sides of the street saying do
03:58not enter
03:58and it'll go right down that street because it doesn't see those signs.
04:01Even after paying for two Super Bowl ads, Dan said he was shocked at the response.
04:1590% agree that this should be banned immediately. Why does Mitza allow Tesla full self-driving?
04:22And it did make a nice splash, but Tesla didn't do a thing.
04:26Well, that's not entirely true. The company did send the Dawn Project a cease and desist letter.
04:31Dan thinks that people should boycott Tesla until the company proves that their full self-driving
04:37software is safe. Pull the product off the market, FSD. That doesn't mean pull the Teslas off the
04:42market. You can still drive your car. You assign 10 engineers or 100 engineers or whatever it takes
04:47to do it and you fix that problem and now you can go back on the road. Before the Tesla
04:52stans jump into
04:53the comments and say, oh, version 14 of FSD is so much better. Well, the Dawn Project tested the latest
05:00version too. And guess what? The Teslas are still passing school buses and ignoring the school
05:06speed limit zones. Just to be clear, they have made improvements. One out of every five times,
05:12it will not stop. But it's not just Tesla. Waymo is under investigation in Austin, Texas,
05:18for passing school buses. And in Santa Monica, a Waymo hit a nine-year-old outside of a school
05:24during dismissal. Thankfully, the child wasn't seriously injured. I was holding off on explaining
05:30this until now, but for the rest of the video, there's a concept you need to understand.
05:35The SAE six levels of automation. Let's start with level zero. Level zero is a car that has no
05:42automation. That's most of the cars on the road today. Level one automation is when the car can steer
05:48and brake on its own, but not both at the same time. So that's kind of like the lane assist
05:54in my
05:54mom's Honda CR-V. Wow. I really do think it is. It's the first time you used it? I didn't
06:01even know
06:01I had it. Level two sounds simple, but it actually is more complicated. And it's where most of today's
06:08self-driving hype lives. Level two is when the car can steer and brake on its own, but it requires
06:15the
06:15human to still be paying attention to the road. And even though it seems way more capable, Tesla FSD
06:21is still only a level two driver assist program. Oh, it gave me a warning that I took my eyes
06:28off
06:28the road for too long. And that allows them to evade all the robotaxi regulations because the
06:34robotaxi regulations only apply when you say that you're level three or above. But here's the thing
06:40about level three. It comes with what engineers call the handoff problem. It means that it's
06:45automated until suddenly it calls for human assistance and you need to step in right away.
06:50Level four is where things start to look autonomous. There's nobody in the driver's seat. And yes, Waymo
06:56is level four most of the time. But sometimes they get stuck. When they get stuck, Waymo says,
07:02well, there's just sort of phone a friend. And that friend helps the robotaxi complete the driving task,
07:09which makes them partially a driver. Level five means that the car can drive itself anywhere at any
07:16time and under any circumstances. It's the holy grail of autonomy, but it doesn't exist yet. And
07:22there's another thing that's keeping robotaxis from being fully autonomous, mother nature. Not a lot of
07:28people on the roads today. It's one of the worst snowstorms we've had in New York in a couple years
07:34now. But gotta go clean off my car. Comes with the territory, you know. Storm like this, when you get
07:40like a foot of snow, you gotta clean off your car a couple times. It's so cold. It's 19 degrees.
07:50For cameras and sensors, snow is especially problematic. It can erase lines on the road and
07:55make everything look like a blank white sheet. Plus, those snowflakes can create noise inside the lidar
08:01and radar sensors and even block the cameras. Waymo says their cars are already driving year-round
08:07in snowy conditions in 10 states, including New York, Michigan, Colorado, and Minnesota,
08:13in addition to earlier testing in California. The Jaguar I-Pace Waymo's already have sensor cleaning
08:19technology, and the upcoming Ojai model will build on that with improved cleaning
08:24technology for heavier weather conditions. Are you engineering for snow now? Tell me more.
08:29Talk more about that. That's happening in the next release, right? The newer version of the car,
08:33right? The Ojai, you know, is going to have, you know, cleaning technology that's better than
08:38than our current version. And so all of that, you know, gets us ready for driving in heavy snow.
08:47On the other hand, you can see a lot of videos online of people using full self-driving in their
08:51Tesla
08:52in snowy conditions like this one, and they say it handles pretty well. And I am just so happy right
08:57now that FSD is driving this way in this snow. Waymo admits that winter weather scenarios are
09:04underrepresented in their training models. Now, in order to make up for that training when they don't
09:10have cars in scenarios like this, they've added to their simulated models to help train the cars to get
09:17better at dealing with winter weather. Those simulations can also be used to deal with what
09:23engineers call the long tail problem. Things like encountering wet cement or a car hauling a tree
09:30that's swaying in the back or a man in a wheelchair chasing a duck across the road. Those are all
09:36real
09:36life things that actually happened. When I met with Chris from Waymo, he told me that the company's
09:42recently unveiled world model can help tackle these rare safety cases. Transform scenarios from a dry
09:49condition to a snowy condition, as an example, and this allows you to multiply the data. And what we
09:55find is that while there are many, many long tail scenarios, we have a corpus of data that actually
10:00has allowed us to reach levels of performance that are showing up in the result. That was a Jaguar
10:07followed by the new Ojai model. Now that Waymo world model can help the cars learn from situations
10:14that the cars may never actually encounter in real life. And some of the demonstrations they have on
10:19their website include driving into a tornado and dealing with a charging elephant in the middle of
10:24the road. The catch with the simulation is if the simulation doesn't have something in it,
10:30then you'll never train against it. As a really simple example, if your simulation doesn't have any moose in
10:36it, the first time you see a moose, you're going to have an edge case you're not trained for. So
10:41simulations can help. But ultimately, if the edge case is not in the simulation, you have not really
10:48solved the problem. So by some estimates, before they can truly say that these cars are safer than
10:53human drivers, robo taxis should expect to clock billions or maybe trillions of miles so they can see
11:00as many of these edge cases as possible. Humans are not nearly as dangerous as the the self-driving
11:06car industry wants us to believe. It's actually pretty good. We have a lot of fatalities because
11:11there are trillions of miles driven. It's just a numbers game. But there's no amount of computing
11:16power that can predict how the public will react when something goes wrong. And that's why we're headed
11:21to the Mission District. I'm on 16th Street in the Mission District outside Randa's Market. I wanted to
11:28take you here because this is the location of one of the saddest stories in the rolling out of Waymo
11:34so
11:34far. So Randa's Market, there was a cat here named Kit Kat who was adopted by the owner and lived
11:41here and
11:42was known as the mayor of 16th Street. Everybody loved this bodega cat. We talked to some of the guys
11:47who
11:47hang out on this block and they told me some nice stories about the cat. We would roll around over
11:52here get on his back and you know yeah on the stomach and everything and yeah yeah all the homies
11:58man
11:58all the homies. Yeah so what was what was your reaction when you heard the news? I didn't even cry
12:04I was just really mad. The story goes at 11 30 at night after the shop was closed up the
12:09cat was here on
12:09the curb. A Waymo pulled up to drop someone off and right as the Waymo was pulling away the cat
12:15darted under
12:16the car. A female bystander saw this and tried to coax the cat out but the Waymo started to drive
12:22off
12:23killing Kit Kat. In this case a human driver might have seen the woman searching under the car and
12:28investigated before pulling away. It led to a big public outcry there were protesters there was a
12:36media conference here with some elected officials they now sell memorabilia merchandise I paid 25 bucks
12:43for this hat remember Kit Kat there's a shrine here and why is this important obviously Kit Kat wasn't
12:49the first animal killed by a self-driving car and won't be the last but you can see how when
12:54you
12:54have something as beloved as this bodega cat it really sets people off and it brings negative
12:59attention onto this kind of robo taxi coming through their neighborhood. Every time I see a Waymo I spit on
13:05it. For real. I'm not gonna lie. We are gonna be meeting with Waymo this week we're going to one
13:10of their
13:10lots we're interviewing some of their executives if you could say a message to them what would you say
13:14to those people? Do they think they took responsibility for what happened? Right right did they feel like
13:22they took responsibility? Yeah yeah in other words they have a long way to go before they earn the trust
13:28of the people. Oh yeah definitely. Kit Kat remains a cautionary tale but a single accident can bring down
13:35a multi-billion dollar fleet and in San Francisco that already happened. Cruz which was operated by
13:42GM suspended operations in 2023 after one of its cars dragged a woman underneath it she was seriously
13:49injured and that pretty much was the death knell for that service in the city. So I have the incident
13:53report here from the San Francisco Fire Department they utilized the spreaders and cribbing to remove the
13:59individual and she was later transported code 3 to San Francisco General Hospital with significant trauma.
14:07Right now I'm under an underpass and I'm next to a parking lot but why am I here? Look through
14:14here.
14:15You see that?
14:19This is what a robotaxi graveyard looks like.
14:26The cars in there are literally sitting around just gathering dust and
14:30grime. You would have thought that maybe they would have moved them somewhere else.
14:36Between Kit Kat and Cruz it's clear that self-driving cabs pose a risk to both pedestrians and pets.
14:43And even if that safety record improves none of these companies can yet answer the trillion dollar question.
14:50How many people does it take to drive a robotaxi?
14:54Congress pressed Waymo on where its remote assistance workers are located.
14:58Waymo says they had already disclosed this to regulators in the press,
15:02but the hearing was the first time many in the public had heard it.
15:06In what countries are these employees located?
15:09The Philippines.
15:11Jimmy?
15:11The Philippines.
15:12So they are in the Philippines.
15:15Yeah.
15:15Mr. Pena, that is completely unacceptable. And here's why.
15:19Waymo revealed they have around 70 remote assistance agents on duty.
15:24Around half of them are overseas.
15:26So if that remote person is in the Philippines,
15:29you have some issues of jurisdiction and accountability.
15:33But whether it's 70 workers or 700, I had a personal experience that tested the limits of
15:39what Waymo's remote assistants are able to do.
15:42I was supposed to go pick up my colleague Lloyd,
15:44and then head to Mountain View to interview executives at Waymo HQ.
15:48He lives in an area with terrible cell phone service.
15:52So when the Waymo passed his house, I had no way to tell it to turn back around.
15:56I still don't have service.
16:01And the app is not working.
16:04Okay.
16:08Now the app looks like this.
16:10This turned into a 30-minute ordeal with a remote assistance team.
16:14Even if you tried turning off and opening your data as well, it's not really working?
16:20Uh, I'll try that.
16:22Okay.
16:23Just let me know if there's still an issue, okay?
16:27Okay.
16:27They were unable to get the trip started again,
16:29and I had to wait for my co-worker Lloyd to show up and save the day.
16:33Lloyd, I am...
16:35Dude, I told it to put...
16:37Okay.
16:38First of all, cell phone reception in your neighborhood is not great.
16:42Lloyd, help me!
16:43Lloyd!
16:45Help me, man!
16:47Finally, Lloyd, who had cell reception, was able to order a new robo-taxi,
16:52and we continued on our way to Waymo HQ in Mountain View.
16:56Nice cardigan.
16:59Bye-bye, Waymo!
17:01All right, let's see.
17:04Lloyd Lee writes for Business Insider about self-driving vehicles.
17:09Really? The Smiths?
17:11In terms of, like, how many humans are operating each robo-taxi, it's kind of hazy,
17:14because you have the people in the call centers, the remote support agents,
17:18but you also have people at the depots who have to charge and maintain these robo-taxis.
17:23And then, of course, there's the engineers who are sort of supporting the whole robo-taxi push.
17:28In simple terms, why does it matter how often a human has to intervene in these rides?
17:33I feel like it shows the maturity of the technology, or another way to put it,
17:39um, you know, how autonomous are your autonomous cars if you constantly need a human holding its hand.
17:44That was made brutally clear in December 2025, when a major blackout in San Francisco showed how fragile current AV
17:52systems are.
17:53More than 800 Waymo vehicles were in the outage area and experienced over 1,500 stoppages.
18:01Waymo says 96% of those stoppages resolved autonomously, but the remaining vehicles blocked intersections
18:08and created problems for first responders and other drivers.
18:12Something as common as a blackout should have been part of the simulation,
18:15and even if it was, Waymo looked like they weren't prepared.
18:18So today I'm at Waymo's San Fran Depot to see how the humans behind these robo-taxis are keeping them
18:24on the road.
18:25There has to be at least 60, 70, maybe 100 people working here right now.
18:31There's a big crew of people. It looks like there's a bunch of different job functions.
18:36Everyone's wearing yellow work vests, and they're servicing these Waymo's as they come in.
18:42You can see the charging stations here. We've got our gloves, our spray bottles, vacuum cleaners.
18:55It would seem that they go through a ton of towels based on the fact that they have one for
19:02Wednesday,
19:03one for Tuesday, one for Monday. That's Monday's towels. One day of towels. And that's all the dirty towels.
19:11There's different lights, colored lights, on each of the vehicles.
19:14And there's a chart that I guess is to let the workers know what kind of service each one of
19:19those vehicles requires at the time.
19:21You've got things like a biological slash lost item, which I'm guessing is like, you know, food,
19:27or maybe someone threw up in the back, or maybe there's, you know, something that what I did, like an
19:33idiot,
19:33is leave their wallet in the back, which, by the way, I was able to retrieve. Thank you, Waymo.
19:38Oh, oh, oh, oh, oh, ah, ah, ah, yes, thank you, Waymo. Thank you, Waymo. Thank you, thank you,
19:50thank you, Waymo. The robotaxis do not have to be fully autonomous for this to make sense as a
19:55solution. The number is never going to be zero. If you have the same number of people involved in the
20:01robotaxi, but they're outside of the car instead of inside of the car, I don't see how you make this
20:06economically viable. I didn't get this conversation on record, but I did meet someone while I was here
20:12who's a Waymo driver. And that means that basically he gets paid to sit in the driver's seat of the
20:19Waymo and either drive manually and his moves, everything he does is recorded and sent back to
20:25train the data or he's driving in autonomous mode and he does what he calls triage. So he uses these
20:34buttons on the steering wheel to send data back to the developers. One thing that stuck to me is that
20:40he told me that there are stretches, especially when you're driving around in autonomous mode that
20:45he starts to drift off. And, um, the seat, the front seat actually notices and buzzes or shakes him
20:52awake, uh, which I thought was pretty funny. So while the future of self-driving cars might look
20:58sleek from the outside, there's humans at every step of the way, driving these things to train them,
21:05tagging data on the backend, cleaning them, servicing them, handling remote operator calls,
21:12and, uh, getting shaken awake when they fall asleep at the wheel. How would you characterize the moment
21:18we're in right now in terms of the deployment of robo taxis, at least in America? People like to
21:24think that we're 99% there because there are hundreds and in one case, a couple thousand robo
21:30taxis out there on public roads driving around. And it feels like we're almost there, that it's almost
21:35solved. It feels like it's 99% done, but I think of it a different way. The difference between having
21:41a
21:41couple hundred robo taxis and a couple million robo taxis is dramatic. There's still many, many issues with
21:48safety and deployment and scalability. And so I think of it as, well, we're done with the first 99%,
21:54but we still have another 99% to go. Now we're up to full speed on the freeway. It's going
21:5965.
22:03And, uh, not bad, not terrible. If it's not clear from the depot, owning and operating a fleet of
22:09vehicles is an astronomical cost. So that's why anybody who's trying to make a robo taxi business work
22:16is trying to reduce those costs as fast as possible. And their newest generation of cars
22:21are actually going to be not the Jaguar I-Pace, which costs an estimated 150K,
22:27but they're going to be based on the Hyundai Ioniq. And that is estimated to be about half the cost.
22:34What remains unclear is how much that cost reduction will help Waymo close the gap in pricing with other
22:41rideshare services like Uber and Lyft. So I downloaded this app. It's called Hackney and, um,
22:47it basically compares the prices between Uber, Lyft and Waymo. The price, the price that came up on the
22:54Waymo is $13.05. Just regular UberX is $23. So it's even twice because of the surge pricing and it's
23:04got
23:05$11.91. So in this case, Lyft would have been the winner. According to analysis by journalist
23:11Om Malik, Waymo needs to be targeting more expensive rides, like $30 a pop, just to make the numbers work
23:19on their current $126 billion valuation. But right now, it seems like Waymo is more in a spaghetti phase.
23:27They're throwing everything out the wall and seeing what sticks. In Nashville, Lyft manages the fleet for
23:32Waymo. And in Dallas, Avis takes that job. In Austin and Atlanta, Waymo rides are booked through the Uber
23:39app. But Uber is responsible for managing the fleet. We've tried to keep all of our options open,
23:44keep all the doors open. And so we're learning a lot right now from the different types of models
23:49that exist in each of these markets and are going to continue to explore as we expand to more cities.
23:55We recently shared that we've been doing deliveries with DoorDash. Waymo may be taking that
24:01partnership literally. They've been paying DoorDashers to show up at stranded Waymo's just to close the
24:07door. I don't think we're at a point in time where we want to say this is the one way
24:11to do it and this
24:11other way is not the way to do it. It's so early in the widespread deployment of these things that
24:17I think it's premature to kind of declare one way or the other. Now Tesla's model is radically
24:23different. They're selling cars to consumers today with the promise that in the future,
24:28they'll become self-driving cabs that can earn money for the owners when they're not using them.
24:33It'll be like an Airbnb thing. You can add or subtract your car to the fleet whenever you want.
24:39Your car gets added to the fleet and it just makes money for you while you're gone. Yeah.
24:47Markets seem to be bullish on Tesla getting their cyber cab business off the ground. Morgan Stanley
24:52analysts expect one million cyber cabs on the road by 2035. And this chart shows how they're projecting
25:00that robo taxis will take over a third of all rideshare miles in the years to come. But overall
25:07self-driving car use will remain pretty flat. But if robo taxis remain a novelty, how will these
25:13companies deliver on their biggest promise? That their cars are safer than human drivers. Every year,
25:20there are over 40,000 deaths on American roads. That makes this country the most dangerous
25:26developed nation to drive in. That's a death rate of about 12 people per 100,000 population.
25:32There are several explanations for this, including the fact that drivers are more aggressive and
25:37distracted than ever before. The argument goes that robo taxis never drive tired or get drunk,
25:43and that they're already safer than human drivers. And as we see more of them on the road,
25:48they could reduce traffic fatalities to near zero. Waymo does have a pretty good record so far.
25:54At least right now, on par with Uber and Lyft drivers, which is to say that they have accidents
26:01about four to six times more often than your average driver, which is an amazing achievement. I want
26:08people to understand. Like, I think Waymo has done a very good job of getting at least on par with
26:15a
26:17more high accident population. But you know, still, they've done a pretty good job. But what if I told
26:24you that some people have already seen enough to say that robo taxis will actually make streets more
26:30dangerous, and that they're willing to go to extreme lengths to prove it? In 2023, as robo taxis
26:37started to take over the streets of San Francisco, a group of activists calling themselves safe street
26:44rebels started staging protests. Basically, they figured out that if you take an orange traffic
26:50cone and put it on the hood of a Waymo, it can stop it dead in its tracks. It was
26:55us showing through a
26:57very passive and peaceful object of just a cone that this technology is not as good as it seems. We
27:06never ever once destroyed a car. We were never vandals. We were also showing that the technology
27:12just in its inherent form didn't work very well. For example, I don't think it happened here, but in LA,
27:18you know, when you had some of these anti-ice protests, people really took their rage out on
27:22Waymo's. I don't know if that was just convenience or if that was like a targeted thing, but I did.
27:27There's some pretty gnarly videos of people just taking rage out on Waymo's. Yeah, that wasn't safe
27:31street rebel. But I do think it it speaks to something that's people are really fed up broadly
27:37with big tech right now. They're really fed up with the ways that corporations have gotten away
27:42with so much. To see how many times Waymo has injured someone, injured someone's pet, injured
27:48someone's loved one, caused havoc, blocked an emergency vehicle. To see all of these times that Waymo has
27:54messed up, we have to ask who who's being prioritized. Is it the driverless car companies or is it the
27:59people? And it really feels like the driverless car companies. And I think people are really fed up
28:03with that. Adding cars to the equation never makes the situation safer. My biggest frustration that I
28:09have heard a lot is they're the future. And it's just like we thought Laserdisc was the future.
28:15Ren and Austin took us on a tour of their hometown to give us a look at the car-free
28:21future they're
28:21hoping to build. Oh, there it goes. Yep. Right past me. This is a slow street right here. The intention
28:31of
28:31the design is that cars only drive to destinations on the block they enter. They don't drive multiple blocks
28:37up and down the street. You can see behind us right here, a set of barriers in the street
28:43that block the incoming lane. Yeah. I mean, you can just see with the sheer number of people
28:48riding past us, these are some of the most alive streets in San Francisco. Yeah. Slow streets work.
28:54They're a very, very effective design of traffic calming, but Waymo's don't act like humans and
28:59traffic calming is designed for humans. They are not really understanding the barriers. So there can be a
29:05lot of issues. There's one right now. It actually looks like there's one. So this is a Waymo waiting in
29:12the middle of the intersection so that it can get into a slow street and it's giving up. They also
29:19showed us Sunset Dunes, which was a strip of Oceanside Highway that was recently converted to a
29:25car-free park. It's just a wonderful place where everyone can come together and enjoy the beach. There's
29:31something that a Waymo will never be able to do, that public transit and biking and slow streets do,
29:36which is bring community and get people outside talking to their neighbors. Our slow streets are
29:40some of the liveliest streets in our city where neighbors know each other. They host really wonderful
29:45block parties. If the real goal is fewer people dying on the road, autonomy isn't the only answer.
29:52It might not even be the best or fastest one. Last year, Helsinki, Finland announced something remarkable.
29:58Zero traffic deaths. Not zero pedestrian deaths. Zero traffic deaths, period. They didn't get there
30:05with self-driving cars. They didn't wait for AI to solve human behavior. They did it by slowing cars down.
30:12City officials lowered speed limits across residential areas and the city center, enforced them with speed
30:17cameras, and redesigned streets to make driving feel less comfortable. Roads were narrowed. Trees were
30:24planted closer to traffic. The city invested heavily in public transit, cycling infrastructure, and walkable
30:30streets. The logic was simple. Slower cars means fewer deaths, and data backs that up. Drop a car's
30:37impact speed from 40 kilometers per hour to 30, and the risk of killing a pedestrian is cut in half.
30:43And here's a shocker. Half of traffic fatalities are the 9% of people who still don't buckle up. What
30:50if we
30:50just make seat belts great again? As we pour billions into teaching cars how to drive like humans,
30:55places like Helsinki are quietly asking a different question. What if we just stop designing cities that
31:01let cars kill people in the first place? And that may end up being the hardest problem autonomy can't solve.
31:09So what happens next? Waymo keeps expanding. Tesla keeps shipping new versions of full self-driving.
31:16And to be fair, there is real progress. In early 2025, Waymo reported zero at-fault
31:23injury crashes across more than seven million fully driverless miles. There are asterisks to that safety
31:29record. And the reason that you're seeing very, very low highway miles out of Waymo is because this
31:35is where it's really dangerous. When you're going 65 miles an hour, this is when you can kill someone.
31:40So I do think that highway driving is going to continue to be an albatross around their neck,
31:46as is scaling. I think Waymo is the more that they try to increase the fleet size,
31:51the holes in the dikes are really starting to show. Right now, what we see is Tesla is not
31:58yet even into scale up mode. They're still trying to get their technology reliable enough that they can
32:03start to scale without human drivers involved. This is not an overnight success story. This has always been
32:09a multi-decade slog. And now we're a decade or two in, and there's another decade or two to go.
32:15So if Waymo's remain a city-by-city grind, and Tesla's remain a driver-assist system,
32:21where will we see full level five drive anywhere, anytime autonomy first? Probably somewhere like
32:28the port you hear behind me. Oakland has one of the largest ports in the country,
32:36and we're standing here. You can see a constant stream of large trucks and shipping containers
32:42getting unloaded from cargo ships, coming and going all day, every day. This is one of the busiest ports
32:49in America. I expect it's much easier to succeed, and we've already seen successes in things like
32:55transportation-enclosed environments like ports or logistics hubs, things like this.
32:59Those are the easy pickings, and that's where it's much easier to have success.
33:04You could just see these epic lines of trucks, and every single one of these guys is in their truck
33:11because they're moving up probably 50, 100 feet every few minutes. And just imagine a world where
33:16they could leave their trucks and maybe, you know, go take a break or whatever, and it would still
33:22move itself into line into the loading zone. That's kind of the dream of what autonomy would look like
33:29in a place like this. I just felt like the RoboTaxi application at the time didn't tick all the boxes.
33:37It wasn't as compelling of an opportunity. Kodiak CEO Don Burnett worked on Google self-driving cars.
33:43He didn't leave because the tech stopped improving. He left because trucking offered
33:48a more realistic path to real-world autonomy. Kodiak's approach reflects where most experts
33:53think autonomy will arrive first. Industrial sites like the Permian Basin, where Kodiak already runs
34:00highly autonomous operations. We work with a company called Atlas Energy. They're a sand mining company
34:06that provides fracking sand for the oil and gas wells. The misconception about the Permian is that
34:12it's actually remote and therefore there's no traffic. These roads are heavily trafficked.
34:16There's a tremendous amount of trucks going in and out, delivering supplies, delivering sand,
34:21delivering fuel, delivering water, delivering electricity. And by the way, when I say these
34:26trucks are driverless, they truly have nobody in the cab. There's no observer and we don't require
34:31remote monitoring. So they are truly on their own out there. We fully trust them to do all the tasks
34:37that
34:37a human would normally do in the truck, delivering across that entire, you know, multi-hour low delivery
34:44back and forth. In mining, in the closed industrial environments, level five is a thing. And it's been
34:51a thing for a long time. Look, it's the perfect setting. Dull, dirty, dangerous. You can control the
34:59environment. You can control the people working in that environment. Rio Tento is the company I've worked
35:05with directly, but many others have fallen suit since then. Automation, at least in terms of
35:11self-driving trucks and things like that, it's kind of old hat in your industry already.
35:17Yeah, we've been on an autonomous journey for over 15 years at Rio Tento, where we started off with
35:24building the operations center, which is right behind me here, where there's 600 people controlling
35:30the vast array of activity that occurs 1500 kilometres away in the Pilbara. The first technology
35:38that we implemented was autonomous trucks. The way we control activity here in the control room
35:44does vary from activity to activity. For example, our autonomous drills can be operated of one person
35:52up to eight drills. For autonomous trucks, one person can operate up to 20 trucks.
36:01Despite the concerns and criticisms, the expansion of robotaxis seems inevitable. I even saw Waymo in
36:08lower Manhattan. The company confirmed this car was driving autonomously with a train specialist behind
36:14the wheel. This is the first time I'm seeing a Waymo in New York City. This is crazy. Look at
36:19this.
36:21It's here, but it's got a guy behind it. It's coming. It's coming to New York City. This is crazy.
36:28There it goes, right outside our office. After spending about six or seven hours total in Waymo's,
36:35my experience is that the Waymo driver feels like someone who's just learning how to drive. A little
36:40too hesitant, a little too heavy on the brake, and definitely making some decisions that I found to be
36:45questionable. Oh my gosh. It's like driving with somebody who learned how to drive a month ago.
36:51People cut these cars off a lot because they know it's not a person, so they don't give a
36:55and then it just like it like jerk, jerk, jerk. You know, I'm definitely gonna get motion sickness
36:59before the end of this video. I'm gonna barf. And every single time it does something like that,
37:05if I had an Uber or Lyft driver who did that, I would be giving them a one or two
37:09star rating. I'm sorry.
37:10When I took a test drive of Tesla full self-driving, it just felt way less hesitant and more like
37:16a
37:16human driver. You know, there's a couple moments in everyone's life when they interact with a piece
37:23of technology that just absolutely blows their mind. I can't believe what we experienced just now and
37:31gonna need to take a minute to process this. So are AVs actually safer than human drivers?
37:37That may be the wrong question. When we say somebody, someone just safer than a human,
37:43right? It's a very weak argument for the general public. High level statistics about safety are
37:48just not how people judge these vehicles. They judge them incident by incident. And every weird maneuver,
37:55every viral video, and every news story about a robotaxi doing something it's not supposed to do
38:01costs the industry more than any high level safety statistic can buy back.
38:06If you're selling on perfection, every picture makes you look bad.
38:09So here's a different pitch for robotaxis that Phil thinks might actually work.
38:13We try and make the fewest mistakes possible. We're transparent that we fixed it. And when we fixed
38:17it, it really stays fixed. And by the way, we're not gonna hide anything. You can trust us to be
38:26transparent, to be good road users. Nobody's perfect.
38:29Not perfect, but always improving. Then you can market robotaxis to users who have already expressed
38:35high interest in them. Parents who are too busy to schlep their kids to all the different places they
38:40need to go. And people who just never want to deal with another person in the car. And here is
38:46where
38:46I'll leave you. It's important to keep in mind that many companies have already tried and failed
38:51in the robotaxi endeavor. Argo AI, the startup from Ford and VW, shut down. Apple quietly scrapped
38:58their self-driving car program after spending billions of dollars on it. And Lyft and Uber sold
39:04off their autonomous vehicle units. But those rideshare players are reinvesting in the technology,
39:09and more companies are entering the space. Zoox, owned by Amazon, is already operating in Las Vegas
39:15and San Francisco. Tesla already has cybercabs rolling off the factory floor. Lucid unveiled
39:21their robotaxi concept, and Nvidia wants to build the technology stack behind all of it. It's more
39:27peaceful than humans are driving. These robotaxis will continue to be a mix of impressive, frustrating,
39:33and potentially dangerous. But whatever happens next, whenever you see a self-driving car on the road,
39:40know that right now, someone, somewhere, is behind the wheel.
39:46Hey guys, thanks so much for watching this video. And if you enjoyed this, we've got four other episodes
39:52of The Limit on Business Insider's YouTube page. We did one about why it's so hard to bring back
39:56supersonic planes. What's the maximum weight that a human being can lift? We did one on why living in space
40:03is
40:03almost impossible. And you can also check out our video about why a mile-high skyscraper is almost
40:08impossible. Thanks again for watching.
Comments