Waymo Robotaxis Offer a Glimpse of the Future of Driving
In 1993, science fiction author William Gibson famously said, “The future is already here—it’s just not very evenly distributed.” Gibson’s quote applies perfectly to Waymo’s robotaxis: self-driving cars that ferry you around like a driverless Uber or Lyft.
In fact, Waymo, owned by Google parent Alphabet, is barely distributed at all right now. You can only hail a Waymo in Phoenix (where it provides service across 315 square miles), San Francisco (55 square miles), and Los Angeles (80 square miles), with Atlanta, Austin, and Miami coming soon. Even in its markets, Waymo currently operates only on city streets, not freeways, limiting its ability to handle many longer or commonly used routes.
Riding with Waymo
Evenly distributed or not, Waymo offers a clear view of the future of driving. Tonya and I spent part of the holiday break visiting my sister in the Bay Area, and she treated us and our son Tristan to a pair of Waymo rides in San Francisco that were as much about experiencing the technology as getting around the city. A decade ago, we took our first ride with a now-defunct ridesharing service called Sidecar; I was amused and somewhat chagrined to discover that our article had a roughly similar title and began with the same quote (see “Travelling to the Future, on the Internet,” 24 June 2014).
We’ve come a long way since then, but the overall experience wasn’t too dissimilar, apart from the lack of a driver. Just as we had in 2014, my sister pulled up the Waymo app and asked for a pickup, which took just a few minutes.
We happened to be at a hotel, so the only confusion was that three other Waymos were doing dropoffs and pickups in the same block. Although all Waymos look identical—they’re white Jaguar I-PACE electric SUVs—they have a dome on top that houses the sensor array (including the all-important 360º LiDAR sensors) and displays the initials of the person who hailed the vehicle. Waymo operates about 300 cars in San Francisco and said it averaged about 4300 trips per day in May 2024.
The rides were essentially perfect. The car navigated San Francisco’s hilly and crowded streets with aplomb. At various times, it backed up to let an SUV in front of us back into a parking spot, paused at an intersection to let a jaywalker finish crossing, and correctly avoided a bike messenger swerving in and out of parked cars.
The app experience was as expected and much like using Uber or Lyft, albeit with buttons that let us control the music and ask the car to pull over. We didn’t try the latter, but I imagine it’s so people feel like they can always get out if necessary.
During the rides, we were agog, chattering about how it was fascinating to watch the wheel turn on its own, how it turned the wipers on for us since it didn’t need to see out the windshield, and how it dealt with each slightly unusual traffic situation. We also enjoyed the car’s screens, which showed our route along with real-time representations of the vehicles and pedestrians surrounding it. Thanks to LiDAR, the car could discern far more about what was happening around us than we could. I’ve driven in San Francisco a handful of times and would have found navigating the traffic conditions somewhat stressful.
The trips to and from where we had parked our car cost $11 and $17; the difference was due to surge pricing for the second trip. A comparable Uber or Lyft ride would have been priced similarly.
Waymo Safety
Of course, Waymo is not perfect, and there have been well-publicized mistakes, such as the Waymo that drove in circles in a parking lot for a few minutes (though I wonder why the guy didn’t tap the Pull Over button) and Washington Post tech columnist Geoffrey Fowler’s complaint about Waymos not stopping for him in a crosswalk.
However, these missteps highlight an important advantage of autonomous systems: once Waymo fixes the bug that caused the parking lot circling or tweaks the system to do better with crosswalks, the entire fleet benefits from those improvements. If only teenagers could be updated so effortlessly!
It’s already doing much better than humans. A study by the global reinsurance company SwissRe examined Waymo’s road incidents across the 25 million miles it has driven and compared the number of incidents that could have resulted in a liability claim against the rate of claims by human drivers in the same cities. Waymo had an 88% reduction in property damage claims and a 92% reduction in bodily injury claims—it was involved in just nine property damage claims and two bodily injury claims, one of which was caused by a human driver who was fleeing police, ran a red light, and hit the Waymo, another car, and a pedestrian.
Waymo is expanding slowly and cautiously, probably as much from the worry about bad PR as the need to learn new environments and situations. I’m looking forward to seeing where we’re at in another decade. With luck, the technology will be far more evenly distributed, including in places with lousy winter weather.
In the meantime, if you get a chance to use Waymo in Phoenix, San Francisco, or Los Angeles, I strongly encourage you to do so. It’s magical.
2 family members have used Waymo here in Scottsdale/Phoenix in the last few months, including airport pickups and dropoffs. Both found the service excellent, slightly slower but (for now) cheaper than the alternatives, and they thought it actually felt safer than a human driver. I suspect we will be using them regularly in the future.
When they make mistakes, they roll out to the whole fleet, too! As with much in technology and, particularly, machine learning, the last part of getting to good enough may never be within reach because not everything improves on a curve or is solvable.
However, given my spouse and other family members have driving issues (some at night), I would love safe, affordable autonomous cabs. What Uber revealed and exacerbated is that drivers for hail or reservation are crushed in terms of making a living. Lots of different causes, but the current reality is that while taxi and gig-hailing makes revenue, it doesn’t make a survivable living for most drivers. It keeps many barely afloat.
So there’s a very human thing here compared to generative AI for art, writing, and business: it’s probably better for humanity that less people are put in a position of driving. If people were making a good living and the cost were reasonable for most people, I’d have a different attitude, for sure. But I think the middle-class cab drive is a thing of the past and most Uber drivers are under the thumb of the aglorithm.
I easily get motion sick so I have a difficult time comprehending how excited people get about being driven around! I could never do that.
Diane
There was just a report of a Waymo vehicle constantly circling a parking lot at the airport and refusing to let the passenger out. The passenger reached out to several people at Waymo while being driven in circles before someone took control of the car and stopped this drive from hell.
Good news, Waymo said he wasn’t charged for the trip.
We’ve been commuting with Waymo weekly in San Francisco for a couple of years now, so we have had many opportunities to compare the service with Uber and Lyft. In identical trips at the same time of day, from/to the same locations, Waymo is cheaper (beginning with the fact that you never need to tip the driver).
I won’t go into the sterling safety record, since you have already done that, Adam. But I will point out that the acceleration, turns and stops are far smoother than the capabilities of any human. Best driver ever!
The depictions of surroundings are so accurate that you can guess the breed of the dogs being animated on the screen in front of you.
One night, we had just turned into the mouth of a narrow one-way street, when the car stopped. In front, a broken limb was hanging from a Monterey Cypress, nearly down to the pavement. A human driver would have proceeded, taking the paint damage to the shop the next day. The Waymo’s dilemma was this: If it drove forward, it could strike an object; if it reversed, it would violate a traffic law. For a minute, it did both. Forward a foot, backward a foot, forward, backward, “sister, daughter, sister, daughter.”
Eventually, it chose the possibility of a sentence to traffic school and reversed out of the situation.
Every trip brings a smile.
(The name is problematic. Every clever out-of-towner thinks he is the first to make the same lame pun.)
I linked to the guy’s video in the article. There’s no way to know for sure, but it felt to me like the guy was playing to social media. There’s a prominent “Pull Over” button that he never mentions having tried.
Yes! It was shockingly good. @dianed143, I can’t say for sure that it would be OK for you, but my sister also gets motion sickness easily, and she was fine, even sitting in the back.
I tell people that in the future, people won’t own a car, they’ll just summon a driverless car when they need it. But no one I tell this to thinks it is a good thing.
From a technical standpoint, I’m impressed, but if I’m honest, I have no desire to see any of the current Big Tech companies shooting off a new tentacle. As we’re seeing on a daily basis, they’re all terrible in myriad ways - Apple less than most, but still so disappointing - and most need to be some combination of broken up and heavily regulated.
I remember how optimistic I was in the 90s and well into the 2000s about the potential of technology and the internet to improve the world, but they’ve now been used to break the world in ways we haven’t seen in nearly a century, with no sign that they have any interest in being better (quite the opposite, actually).
If Waymo ever becomes a viable business, I hope that Alphabet is forced to divest and/or license the technology at a fair price.
Several months ago, I had a negative encounter with a Waymo while driving in San Francisco.
Many residential streets in San Francisco are only wide enough for parking on one side plus one lane. They are not wide enough for cars going in opposite directions to pass without one car pulling into the space created by a driveway or red zone. As I was driving to a house on such a street, I saw a Waymo approaching me and pulled to the right into a gap to leave room for the Waymo to pass. The Waymo stopped in place, and it was unclear if there was enough room to pass. Another driver would have either pulled forward into a gap or just proceeded past me. However, the Waymo would not move. Since we had reached an impasse, I went forward and cleared the Waymo by a matter of inches.
Reviewing the incident, I think I should have clarified my intentions by activating my 4-way flasher and, if necessary, turning off the ignition and stepping out of the car.
Hopefully, Waymo has improved its software to handle these situations better since streets like these are quite common in San Francisco and other older cities.
Well, here is someone who does think it is a good thing
I’m probably a safer driver than I was when I was young due to being more cautious and experienced, but I’m certainly not as good of a technical driver—notably, my vision isn’t as good and my reactions aren’t as quick. I anticipate age causing a further reduction in my physical capabilities such that at some point, caution and experience will no longer outweigh my physical limitations in making me a safe driver—I watched that happen with my grandparents to the point where they were forced to stop driving and rely on others for rides everywhere. I fervently hope that we have true robotaxis by then.
And while we’re no longer in this stage of life, having robotaxis available for shuttling kids around would have been fabulous 10-15 years ago. So much parental driving! I do think robotaxis will have to become commonplace before parents trust their kids to them, but I imagine most parents would be more comfortable with a robotaxi than a random Uber/Lyft driver. And of course, it will be appropriate only for kids of a certain age who can easily communicate on their own (ie, they have their own phones). I’m sure much digital ink will be spilled over issues surrounding minors and self-driving cars.
So yeah, I think it’s a very good thing.
It looks like you can report incidents like that to Waymo.
https://support.google.com/waymo/contact/feedback
Here is someone who doesn’t…
I’m not a fan of self driving at all…there are still way too many problems and issues with the tech at this point. I’m not going to claim it will never get there…but even the best AI drivers these days still have issues. We read about Tesla FSD craning far too often still…and just yesterday there was a story about one of the, or cling the airport over and over. Yes…there was a pull over button that that guy didn’t use…but there ar just way too many situations that the AI programmers haven’t thought about but a human driver would know what to do. Not sure it was in this thread or I saw it someplace it…but the situation of a 1 way street and the branch hanging down causing forward/back repeatedly is another example. A human would either just back up, go pull the branch down, or see if it could be avoided and not scratch the paint.
Update: Zoox is now Zooming passengers on public streets.
I continue to take every opportunity to travel in an autonomous vehicle. I had some great rides with Cruise, and look forward to my first Zoox in the coming weeks.
Some observations:
The Cubes are not nearly as numerous as Waymo Jags (yet), but you can’t drive anywhere downtown without seeing a Zoox box or its mapper car. They’re suddenly everywhere.
The mapper cars are not boxes, just ordinary cars dolled up with sensors. However, Amazon chose to paint them all in a charming rainbow.
For now, Zoox appears to be only a corporate shuttle, driving employees between campuses/offices.
I expect the app to appear soon, but when you search for Zoox today, App Store returns this list: Waymo, Uber, Lyft, and…Turo. Yes, Turo.
I believe that Cruise will be revived by a company that buys it from GM.
You know that a new technology has arrived when your spell checker recognizes it. Today it knows Waymo, but you need to right-click Zoox and choose “Learn Spelling”.
A closing note: Anti-driverless anecdotes are tiresome. “I heard where Waymo had this incident… I read that a Cruise injured somebody… My wife’s cousin’s boyfriend got into a Zoox and there was a spider on the floor!”
To borrow a slogan from an 80s social movement: “We steer, we’re here, get used to it!”
But these anecdotes are interesting only because a computer is making them and because there are so few. (Self-driving cars that make a lot of mistakes don’t make them in public.)
We’re completely inured to the vast number of mistakes that human drivers make. But I seldom drive for any distance before I see a person do something that’s not ideal, unsafe, or actively illegal.
I think Waymo (et al) is demonstrating that a computer-driven car operating in a relatively predictable, pre-mapped environment is at least as reliable as a human-driven car. But whether it’s fair to extrapolate this success onto a bigger stage as “the future of driving” remains to be seen.
The current approach does feel like a dead end to me. It relies on an intensive specifying exercise that requires constant monitoring and adjustment for road closures, changes, etc. I can’t see it ever scaling beyond very regular grid cities. Maybe I’m wrong, but so far no one seems to be taking an approach that might have a chance of generalising, so I don’t have much hope that these self-driving taxis will spread that far (or end up being cheaper than human drivers).
Of course in cities it would be much better if all this investment went into a good public transport system (potentially autonomous as in some places) and walking/cycling infrastructure. That’s a much more efficient and equitable way of moving people for these types of journeys. Where I live there’s a good bus network and it provides the freedom for the young, old, and others who can’t drive that Adam talks about above (along with taxis of course). I don’t have to ferry my son around because he can get around on his own.
I realise this might sound like I’m against autonomous driving, but I’m not at all. I think transport technology desperately needs to move forward and get away from how inefficient it is now. But the capabilities and focus of the current autonomous taxis seems to entrench current inefficiencies and inequalities without any hope of moving to a more useful model. And this is to say nothing of the fact that they aren’t being built with privacy as a foundation so using them allows some of the most rapacious companies around to track your every movement.
While I realize you’re in London, which has some wildly squirrely driving, San Francisco isn’t exactly a regular grid city either.
But realistically, Waymo and other companies are absolutely building systems that generalize and scale. Every car contributes sensor data back to the system so other cars can use it, and the cars are designed to deal with on-the-ground situations. They’re far from perfect, but the rate of improvement is significant.
Also, Chinese firms are investing heavily in autonomous vehicles. Baidu is operating now in 11 cities in China, and 54 cities have pilot projects.
While there’s always a role for public transport, Brad Templeton, who does a lot of research and writing in the robocar field, has published a number of articles showing that it’s not nearly as efficient or cost-effective as it might seem, largely due to off-peak inefficiencies, labor costs, and subsidies. This Perplexity search has a summary with links to what he’s written.
https://www.perplexity.ai/search/brad-templeton-research-into-p-5AZcB.fjSjy74UiApVzDGw
Sure…human drivers make mistakes or deliberately do the wrong thing…but they can deal with a completely unknown never before seen situation…where the AI can’t because it is out of its programming logic.
Computers today do not have the processing ability of a brain and even “self learning” software is still limited by its programming…for instance the example I made of the branch hanging down in the one way street where the car went forward and back several times. The human would move the branch, go around it, or a8m0ly back out of the one way street. Computers will eventually be powerfully enough to do that…maybe…but will tha5 computer be affordable to mass install in vehicles? Who knows.
I see two themes in this discussion: One is about the capabilities and the safety record of AI-driven vehicles. The other is about whether self-driving cars are the most appropriate solution to transporting people in urban areas to begin with. The former gets more attention — and not just here on TidBITS Talk — but the latter is much more important.
The technology behind self-driving cars is fascinating, and they already have objective benefits over human drivers.
Even though the latter tend to more flexibly adapt to previously unencountered situations, not all drivers behave well in those situations. And in the majority of standard driving operations, AI already is superior: it never drinks or does drugs, it’s never distracted by its passengers, it never has a bad mood, it never fumbles with awfully designed car UIs, etc.
But as soon as you look beyond the technology, it’s painfully clear that these vehicle do not solve the problem of (de-facto and actual) public transportation, they exacerbate it dramatically.
When Lyft and UBER started operating in the Bay Area, the main criticism was about how they would push out licensed cabs. Cabs are much more expensive to operate due to the required licenses, and the number of available licenses is capped.
Only over time did it become clear that many passengers also moved from public transportation to individual rides. The result was considerably more vehicle traffic and more congestion.
And the way in which companies like Waymo or Zoox offer rides is nothing but a non-license cab that happens to have no human driver. So it takes up just as much space as cars already do, and sometimes even more: Volvo even showed a concept for a single-person autonomous vehicle with creature comforts that rival those of the finest first-class airline cabins.
How much more space does it require to transport a few dozen people in such vehicles compared to a bus, subway car, or tram line?
And that’s the real problem: Despite the lofty promises of their providers, cities will become more and more clogged. Affordable, sustainable, and truly scalable public transportation will whither more and more.
To put some hard data behind that claim: In the next couple of years, the two main transportation agencies in San Francisco are facing hundreds of millions of Dollars in deficits and are struggling to keep operating (BART is Facing a Fiscal Cliff, S.F. Muni faces massive $214 million deficit).
Unfortunately, the discussion about the technology of autonomous vehicles mask these underlying systemic challenges quite nicely.
P.S.: My wife and I visited London, England, recently. Their public transportation system of underground and buses is brilliant despite several shortcomings. Missed the Tube? The next one will be here in a few minutes. Missed the bus? Wait for 10 minutes, and there’ll be the next one. And as for cars? If you absolutely want to drive into inner London, you’ll have to pay a “congestion charge” during certain times of day, and it’ll cost you £15 (ca. $18.50). Per day.
All true, and yet London’s traffic speed averages about 9 mph. An excellent public transit system & congestion charging has not actually sped things up.
You’re correct, but your post also actually supports my point!
Traffic on the streets actually averages less than 10 mph. In comparison, the public transportation options, especially the Tube and “even” trains move considerably faster.
In this context, I think it’s fair to claim that adding self-driving cars to the mix would not improve anything. It’s all about moving traffic from lots of highly individualized vehicles which take up a lot of space relative to the size of their “cargo” (humans and otherwise) to fewer vehicles that can accommodate more “cargo” per volume, so to speak. And even more so in cities that have grown over centuries, where it’ll be almost impossible to expand existing roads, let alone add new ones.
That said, though, I wish public transportation agencies would dare apply “disruptive” approaches to how they offer their services to their customers.
Even with a preference of buses, underground trains, etc., the user experience of ordering, tracking, and paying for a Lyft ride still is dramatically more enjoyable and effective than the typical offerings of public transportation.
IMHO, the Clipper Card in the Bay Area is an example of how move in the right direction. Mine is stored in the Wallet on my Apple Watch, and it’s handling is brilliant. Just the fact that I can instantly add funds to it right on the watch, is nothing short of amazing.
No, it doesn’t. First, your point was that having public transit would reduce congestion. It doesn’t. Second, public transit includes buses, and those buses are creeping through the 9 mile an hour roads just as much as cars are. Third, I used to commute on the northern line in London and between the delays and the overcrowding (which meant you had to skip several trains) the commute time was not better.
I’d say that SF Muni and BART’s ridership and revenue problems stem from multiple sources, to which autonomous car services are only a minor, very recent addition. Even if Waymo and Zoox were banned tomorrow (and keep in mind Cruise has already exited the market), Muni and BART would still face a complex and contentious road to stability and sustainability.
I think you mean “The other is about whether cars are the most appropriate solution to transporting people in urban areas to begin with”
And the answer is obviously no. (Rural areas are different, but we should not be encouraging more people to live permanently in remote rural areas.) I will not be taking questions at this time.
Lets restrict this discussion to the tech of self-driving cars and Waymo. Urban planning policy is far far beyond the scope of TidBITS.
If moderators agree, I would see that as a reasonable restriction. However, I see some discussion of local policies as on-topic because the technology is both enabled and affected by legal regulations and governmental subsidies.
For example, if a county banned nighttime operations, there would be no need to train the cars to drive after dark in that locality. Or if autonomous cars were exempted from congestion charges or tolls, hypergrowth could force rapid changes in the scale and scope of the technology.
Dave Shamino is right. While it’s true, of course, that “technology is both enabled and affected by legal regulations and governmental subsidies”, the politics of policy is off-topic.
Some humans may be able to deal well with unknown situations, but that’s far from guaranteed, and many car crashes result from a driver encountering a new-to-them situation and failing to respond correctly.
Of course, humans do regularly navigate such situations correctly as well, but since every situation is new to every driver the first time, that’s putting a lot of faith in random people’s ability to deal and learn for the future.
One of the things that impressed me in our limited Waymo experience was just how well it handled somewhat non-standard situations, like the car in front of us backing up into us to park and a biker swinging in and out of parked cars. I wouldn’t have relished watching a new teenage driver deal with those situations.
In theory, if a Waymo did encounter such a situation, it would ask for remote assistance from a person.
Yeah, my apologies for somewhat introducing that branch. Ironically, it’s probably a much bigger and more complex topic than self-driving cars.
Let’s stay on the tech topic.
Let’s not forget Apple’s extremely disastrous Project Titan:
https://www.bloomberg.com/news/newsletters/2024-03-03/why-was-apple-car-canceled-the-hubris-in-apple-thinking-it-could-outdo-tesla-ltbke5ut
I actually wonder if autonomous vehicles (car/small vans) could help public transportation in the USA sprawling cities by helping close the “last mile barrier”, shuttling people from their suburban homes to a metro or bus stop at a lower cost than having human-operated vehicles do that.
Yes…but the likelihood of an AI cheap enough and powerful enough to be builtin to the car performing better than the human driver is IMO low. I’m sure that a supercomputer powered AI is probably smart enough to be nearly as good as the human…but that’s not what is going into a mass production vehicle. YMMV though.
A funny related thing is one of the mega-hyped uses of 5G cellular, especially mmWave, was for centralized control of autonomous cars. Guess that’s not happening any time soon!
I’m in Edinburgh which has some small twisty roads, but nothing like London! But in any case, I was wrong about the lack of a ‘regular grid’ being an issue – I had mis-remembered. (I was on my phone when writing the previous response and wasn’t able to properly look up the issues I’ve read about previously.) The ‘intensive specifying exercise’ I referred to is properly called localisation mapping. Brad Templeton describes it thusly:
Creating these known HD maps is a big barrier to generalising autonomous driving. There is a good academic survey of mapping issues and techniques here. I’ve not read the whole paper, but the section on mapping gives a good overview. There are attempts to develop methods to automate the creation of localisation maps, but as far as I understand from my reading they are still theoretical or for use at low speed/low consequence (e.g. robot vacuum).
I am far from an expert in the field, but from my limited understanding relying on localisation maps feels like a brute-force method that won’t scale. It seems like another situation where lateral thinking is needed to approach the problem from a completely different way instead of trying to create an electronic human brain (we create and use ‘localisation maps’ on the fly as we drive). Shades of the decades spent developing generalised AI which is always 5–10 years away vs. LLMs.
The difficulty and effort of creating localisation maps and the computational challenge of sensing these details at speed is presumably why these services only operate on city streets.
But medium distance, rural service is where these autonomous taxis could be transformational. Instead they are providing a marginal benefit over existing city cabs. Areas outside cities and towns often have poor or non-existent public transport and taxi fares are high. There would be massive benefit for these areas to have low-cost autonomous vehicles that could provide on-demand transport.
I read through some of Brad Templeton’s work and it provides very useful data for high level policy planning (e.g. if you want to reduce energy and carbon use in US transportation, focusing on public transport is not the right place to start). But it doesn’t generalise to providing a best approach to transport at the local level (which, to be fair, he acknowledges). I find his writings on ‘robotaxis’ mixed. Some things certainly ring true, but there’s a lot of techno-utopian wishful thinking. He is definitely not an impartial observer either, he is heavily embedded and invested in the autonomous vehicle industry.
These are very real issues, but it’s worth noting the following:
A fleet of autonomous vehicles will also have peak inefficiencies as there will need to be enough of them for peak times and something has to be done with this excess capacity in off-peak. Will they go to massive car parks on the edge of towns? Will building these storage spaces be acceptable or politically feasible?
Modern cities couldn’t operate without mass transport simply due to physics – there’s not enough space for everyone to be in their own vehicle (even if driverless cars are microcar-sized).
The subsidies issue is complicated as it varies greatly by locality and all forms of transport are subsidised (but in very different ways!). Importantly for this discussion, ‘robotaxis’ are currently massively subsidised by venture capital. It’s not clear when (or if?) they will be affordable or cost competitive. There are a lot of background costs aside from the technology in the vehicles – including a significant amount of labour.
Like @jochen, my overall concern with autonomous vehicle development is that I think it’s focused on the wrong problems and approach. This matters because of the massive investment and huge opportunity cost. Surprise, surprise the majority of investment is being spent to build what’s most attractive to silicon valley venture capitalists, not what would be beneficial to a wide range of society. We could be developing transformational transport systems but instead we’re spending billions to get automated chauffeur-driven cars (in limited situations).
I wonder how this system deals with bicycles. We have a lot of those in our Dutch city streets. Often on dedicated bicycle lanes, but not always. I think Tesla complained about the bicycles and suggested to take them off the streets, which was not very well received, to say the least. In our inner cities it is common not to own a car and have multiple bicycles instead, for commuting and pleasure.
I’m really not sure what you’re suggesting. Waymo has 25 million miles of autonomous driving. It’s working now and performing better than human drivers now. Yes, it’s in limited areas, but it’s expanding all the time. And that’s just in the US—there’s as much or more happening in China. So this isn’t a speculative “maybe someday we’ll have self-driving cars” situation.
The Waymo we were in dealt fine with bicycles. We didn’t have a full bicycle lane, but that doesn’t feel worse than single bikes acting randomly.
I’m not surprised that Tesla would have problems since their system relies on cameras, not LiDAR, and thus can’t see nearly as well.
Humans can recover from an unfamiliar situation mistake quickly, thanks to which I am here today. I was going up a narrow off-ramp of route 128/95 near my home at night when I saw the headlights of another car going down the off-ramp directly at me. I swerved to the right, instantly realizing that was the only way to get out. There was no Jersey barrier between the on and off ramps there, so the other driver swerved to their right and bounced over the grass onto the onramp. It was all over in a split second, so I can’t reconstruct what happened, but without both drivers doing the right thing, it could have been a head-on collision.
They can, but they often don’t. I’m glad you did.
That’s my point…no matter how many miles of driving the AI has in its database…the cpu cycles and memory limits in whatever is in the car become an issue. All those miles in waymo’s AI development database still have to be downsampled into software that fits in the car at the price point the car needs for the parts.
Adam’s mileage obviously varies…and that’s just fine.
Which is why there is a lot of research into “connected cars” and “connected roadways”, where the car can offload the heavy-duty computing onto cloud services.
Of course, typical Internet latency (100s of ms) is too slow for this to work, so the industry is focusing a lot of R&D effort in “edge computing”, where cloud servers are colocated at cell towers (ideally near the roads), along with “network slicing” to provide guaranteed 5G bandwidth, in order to minimize the communication delay between the servers and the cars.
There’s also work on vehicle-to-vehicle (V2V) communication, where cars (both autonomous and manual) share data with each other in order to optimize algorithms for avoiding collisions and reducing congestion.
Of course, this is a really difficult problem. It requires critical cloud data to roam from server to server as the car moves from cell to cell. But it’s an active field of research for telecommunications.
The future here is really bright, but it’s certainly not here now, and I wouldn’t place any bets on when it will arrive.
See also:
And cell phone connectivity is soooo reliable that it always works brilliantly…not.
Offloading the computing power…and admittedly servers in every cell tower is technically but maybe not economically feasible…but that’s still doesn’t solve the problem of signal dropout…and again, more cell sites can help that…but Murphy says the signal will drop just as the unknown situation presents itself.
I could see self driving being ok on freeways with enough tech tossed at it…but blanketing the US isn’t happening and even blanketing an entire city with sufficient guaranteed, always works cell network seems problematic. The tech has to be all in the car…and while that is probably technically feasible…it isn’t going to be cheap and few buyers will spring for the expensive option…and without a user base the cell c9mpanies aren’t going to wholesale upgrade their systems for it.
Just like putting 100,0 solar panels and windmills in Kansas and building a new grid with sufficient capability to move ll that power to CA and NY is technically feasible…it is many, many billions of dollars to build and years before the infrastructure upgrade is complete enough to actually solve the problem.
Waymo has now started driving on freeways and has significantly expanded its footprint in the Bay Area.