We keep asking when driverless cars will "really" be here, but it's an odd question. They already are:
California regulators agreed on Thursday to the expansion of driverless taxi services in San Francisco, despite the safety concerns of local officials and community activists. In a 3-to-1 vote, the California Public Utilities Commission, which regulates self-driving cars in the state, gave Cruise and Waymo permission to offer paid rides anytime during the day throughout the city.
OK, this doesn't mean we have full-fledged driverless cars capable of going anywhere at anytime. But it does mean we have driverless cars capable of driving around the city anywhere at anytime—and if that's not "really" here, I'm not sure what is.
If Waymo and Cruise can pilot their driverless cars reliably around San Francisco, 100% autonomous cars can't be too far away. Highway driving isn't a big deal compared to city driving, and that takes care of all driving on streets. All that's left is non-street driving: parking lots, drive throughs, dirt roads, and so forth. I'm not sure how hard that is to automate, but I'm willing to bet that my longtime prediction of 2025 for honest-to-God driverless cars is pretty close to the mark.
And not interfering with emergency vehicles. And coping when someone puts a traffic cone on the hood. And …
"It's 90% solved"
- the most ominous phrase in engineering
Absolutely true, @royko.
I've been developing engineering products for forty years, and one thing I can say is that it takes 80% of the projected time to get the project 80% done, and 80% of the time to get the next 20% done.
We might be at the 80% point with "driverless" cars right now, and cars have been around for how long?
And lots and lots of products don't work when the project is finished. Most, in fact. I'd like to point out that driverless cars are still regularly killing their passengers.
It is more like a Zeno's paradox. The first 80% takes 80% of the time. The next 80% takes 80% of the time too. I have always believed is that 80% of the remainder though -- or 16% of the original project. No project ever gets truly complete.
This is why you have to download weekly "updates."
Beg your pardon, but don't you mean it takes 20% of the projected time to get the project 80% done? I'm sure it's just a typ0.
The 90-10 rule applies. The first 90% of a project takes the first 90% of the time. The last 10% takes the other 90% of the time, and the 90-10 rule can apply there, as well.
…and dealing with “connectivity issues” during large-scale events: https://www.kron4.com/news/bay-area/video-cruise-cars-halt-traffic-in-sf-cite-connectivity-issues-linked-to-outside-lands-festival/amp/
From a different NYTimes article:
"the agencies concluded from an analysis by the San Francisco County Transportation Authority that self-driving cars, on average, resulted in more injuries than vehicles operated by human drivers. But the C.P.U.C. said in June that the data that city officials had based their analysis on was “problematic” since it excluded incidents involving self-driving cars where the human drivers were at fault."
‐---------
Kevin has defined down success to merely getting government approval. Which doesnt mean that the cars actually work very well given the apparent accident and failure rates.
I doubt that anyone would bet against driverless cars 'arriving' if they never have to solve the basic safety and failure rates before they 'arrive' . 'Arriving' only means getting the government to accept the high failure rates.
This is a political problem, not an engineering problem. Political problems are easy to solve when you have a ton of money and an established lobbying effort.
"Who was at fault" is a legitimate distinction. For instance, rear end collisions are effectively always the fault of the person behind. If self-driving cars get rear ended while sitting at red lights, which is going to happen, that should not be counted against the self-driving car's safety statistics. If the point was reached where ALL crashes involving self-driving cars were the fault of a human, then I would describe self-driving cars as perfectly safe.
What about the Tesla the pulled over to the left lane and stopped suddenly in the middle of the Bay Bridge? I think there's a case to be made that the vehicles behind were not at fault for that.
I don't know the facts of that case, but in general with auto torts, there are exceptions to the rule of thumb that the car that rear-ended the other is the one at fault. Swerving in front of the other car and hitting the brakes is very much one of these exceptions. It can be hard to prove, though. At least traditionally. With driverless cars I expect there would be ample data available. And the exception only applies if the accident is before the front car has established itself in the lane. In other words, you gotta give the guy behind you a chance. What precisely does it mean to say a car has or has not established itself in the lane? That's what juries are for.
That would be why I wrote "effectively all". There is one rear-ender in many thousands that is not the fault of the guy behind. On the other hand, sometimes cars stop in unexpected places; even in the left lane on the freeway you should be prepared for the guy in front of you to pull a sudden panic stop, and you should be far enough behind to stop without hitting him. (I'm aware that in heavy urban traffic, that's impossible, but all that says is that it's impossible to drive safely in heavy traffic.)
Human drivers do stuff that is much stupider than that. A couple of months ago I had to dodge an SUV going backward down a one-way street, apparently because they missed a freeway entrance ramp... by almost a quarter of a mile.
It would be great if driverless cars were perfect from the get-go, but for now I'll settle for better than human drivers.
"If the point was reached where ALL crashes involving self-driving cars were the fault of a human, then I would describe self-driving cars as perfectly safe."
I'm not sure this follows. The famous incident in Arizona where a self driving car just plowed into a jaywalking pedestrian and did not slow down or stop is a case in point. By some metrics, the car was not at fault there. But if self driving cars are going to be safe they need to avoid collisions where they are NOT at fault, just as human drivers strive to do.
The Kevin Drums of the world will be happy as long as the autonomous vehicles don't actually kill anyone more frequently than human-driven ones do. So if we end up in a future where autonomous cars are simply knocking pedestrians on their asses a lot or giving the occupants inside them whiplash from hard breaking/swerving to avoid accidents, it's all good!
I'll be happy with that as a starting point. I saw a guy t-bone a car at an intersection because he was racing to beat a red light (actually, the light was red well before he entered the intersection). If driverless cars are not perfect at first, but still outperform human drivers, that's still a win. And they'll get better.
If all crashes involving autonomous vehicles were a human's fault, it would pretty much have to be true that the autonomous vehicles were also doing a good job of avoiding crashes that would not be their fault.
Useful info that is suspiciously left out of the promotional press and Kevins fawning coverage of driverless cars:
US driven miles in 2022: 3.2 trillion
Waymo/Cruise autonomous miles: 2 million
Us driving deaths: 1.35 deaths per 100 million miles
Expected autonomous driving deaths if they acheived safety identical to human drivers: .027
This calculation wildly overstates the expected average as the auto cars are not using trucks/SUVs and are avoiding highways. These types vehicles and high apeeds are where all the deaths occurr.
That auto cars are reporting 0 deaths is a garbage stat. But the fans will promote it.
I agree with you as far as you have taken this. Deaths aren't a good metric as yet.
However, accident count, and accident count split up by seriousness are also good metrics. They happen a lot more, and we should have good data by now. We do have good data by now. It overwhelmingly tilts in the favor of AVs, which have far fewer, but not zero, accidents per mile.
The simple answer for this is most accidents are caused by inattention by human drivers. Not because of some complex scenario, but because someone looked at the newspaper ads on the seat beside them and didn't notice the red light ahead of them and thus rear-ended the cherished Jeep Cherokee sitting at the light, totaling it.
Yeah, that happened to me. That was my Cherokee. I'm still a bit salty. I wasn't injured, though. I just lost my car, and couldn't buy another one, as it was discontinued by then.
AVs will make mistakes, but not this kind of mistake, which is very common.
How do you know that AV have far fewer accidents or injuries per mile?
San Francisco County concluded the opposite after the AV companies were forced to disclose all accident data.
This is why collision avoidance systems being able to take priority is basically as far as I see AV tech progressing in the broader world. It will address the inattention from a human driver problem, and that's as far as it will go.
No, most accidents are NOT caused by trucks and SUVs on high speed roads. That may be the leading cause of multi-vehicle fatal accidents, but a huge number of auto accidents happen on 25 mph and 35 mph roads, often involving pedestrians (who are absent from high speed roads). And passenger cars are involved in plenty of those.
Where can one find the stats for that? In my search I came across https://www.iihs.org/topics/fatality-statistics/detail/state-by-state which has "Rural" vs "Urban" but that would be at best an imperfect proxy for "Highway" vs "Surface Street"
To have a comparable "expected autonomous driving deaths" don't you need that "Highway" vs "Surface Street" breakdown to filter-out the (trillions of?) highway miles to leave surface street miles by human drivers?
You are not correct. Lots of traffic deaths occur at lower speeds in urban areas, and people in cars, not trucks/SUVs, are very definitely included in the death toll. Highways are actually considerably safer in terms of fatalities per 100 million vehicle miles than city streets are.
It does meant that they aren't running over pedestrians though. On the flip side, I saw a Cruise car cut a human driver off a couple o f weeks ago by changing lanes into their path to avoid a car backing out of a driveway. The human driver was awake enough to avoid a collision. The other point is that the weather in San Francisco is usually really, really good. It doesn't snow, there is almost never any sleet or freezing rain or ice on the roads and it rarely even rains very hard. So what Cruise and Waymo has shown is that they work pretty well in ideal conditions
Kevin appears never to have driven a car anywhere in which there is real winter weather. Snow obscures markings. And most roads in my area in central NY are poorly marked under the most clement of conditions. I don't see how driverless cars will ever work in places such as these. There's a reason they are currently limited to areas in which the roads are almost always in perfect condition.
Although currently driverless cars stay in the "easy" areas, they can have sensors which are way superior to what humans have. In theory a driverless car could see through the snow and locate any objects under it in ways that a human can only guesstimate. Which is not to say competing against human's heuristic guesstimation will be easy to beat but better sensors and perfect can compensate for some of the processing differences between humans and computers.
They can be made to work here...if we completely rethink and rebuild essentially all of our roadway infrastructure. Technically possible. Politically impossible.
I'll toss my hat(s) in and say that 100% driverless cars won't be feasible unless they interact with the road they're on. How much will this raise the cost of laying down new asphalt? In my expert opinion, 'a lot'. Once again, expect the rural part of the country to lag everyone else. Come to that, suburbia as well. Come to that ...
Precisely. Well marked interstates in good driving conditions? Sure. My neighborhood and most other neighborhoods in the NE? LOL
I'm not sure this is better. The way it works in real life is very different. Consider a typical undivided road with two lanes in each direction. The bracing snowfall occurs: not enough to shut things down, but enough that driving is not business as usual. What happens, as the snow accumulates, is that those two lanes in either direction merge into one, typically down the middle, as drivers sensibly avoid the edge of the road and oncoming traffic, and later drivers also sensibly follow in the tire tracks of those who came before them. Then the plows come through, shoving the snow to the right. If there is not much snow and there is a wide enough shoulder, this restores the status quo ante. If either of those conditions are not true, you end up with the right half of the right line under a pile of snow. This reinforces the de facto single lane. Depending on local conditions, this single lane might last weeks.
So when I read about the autonomous vehicle that is sensing the official lane markings, I imagine it perched halfway up a wall of snow. Presumably the carmakers will eventually figure out that this is a problem, but this just brings us to what we should have been talking about in the first place: Can the car figure out the de facto lane when the de jure lane does not apply? I wouldn't think this problem is insurmountable, but in the years I have been following the topic I have never once seen one of these outfits discussing it. It very much has the feel that these things are designed for the mean streets of Mountain View.
If my experience is anything to go by, I’d say they are designed for the mean streets of San Francisco. The streets here can be pretty “mean”, but you are right, no snow.
If you can imagine an autonomous car sensing lane lines under snow, then you should also be able to imagine that it can sense walls of snow and avoid being found half way up them.
Horseless carriages can never work. They're noisy, smelly, dangerous, unreliable, and they frighten the horses.
Is the set of places a horseless carriage can go now a proper superset of those reachable on a horse?
Yeah, that would be nice. I wonder when we'll get there?
Hint: humans behaving like the autonomous deathbots do would instantly be arrested, and we're constantly gaslighted with irrelevant statistics in return. But fuck the pedestrians, we're a car nation and capital says the little people can suck it.
Progress brings a tear to your eye, doesn't it?
Sadly, 'humans behaving like the autonomous deathbots' do not de facto get instantly arrested. Especially in South Florida, according to my best friend who's been a transplant for going on 30 years.
In this sort of project the level of difficulty rises exponentially with the level of success achieved. 100% is unachievable, the approach to it is asymptotic, the closer you get the slower additional progress gets.
Hence: 90% is far from a finished project.
Also: We don't just want to avoid casualties, we have to minimize fender benders as well if this thing is to be scaled.
I still think that this is the dumbest invention ever undertaken: totally unnecessary and yet it diverts resources from essential problems like how to deal with climate change.
40,000 deaths per year is not "totally unnecessary".
It is a milestone.
Though it may just reveal the need to have sensors and guides built into roads to get driver-less cars to work well enough to be useful.
"Highway driving isn't a big deal compared to city driving, and that takes care of all driving on streets." A statement from an suburban resident. What abouut all the secondary rural roads? Those have far more variability than urban and suburban roads. No lines,no curbs, much more variable paving, camber changes, narrow bridges etc. all add to the challenge for the driverless programes. I;m sure they can be solved but the incentive decreases as the consumer base decreases.
There is much talk about snow. I would think the average unmarked gravel road would be just as much of a challenge in summer as any snow covered winter road. I've never heard of autonomous vehicles operating on unpaved roads, although I presume somebody has at least looked at the problem.
Yep - DARPA had some challenges out there for people to compete/complete: https://en.wikipedia.org/wiki/DARPA_Grand_Challenge
Nobody's asking one of the fundamental questions: Do you need to tip the car-bot?
Suddenly we have a situation where the use of an AR-15 is entirely justified. Good to know!
I'm going to push back. Driving in SF isn't that hard. Sure, it's a city with pedestrians, bikers, construction, whatever. But, and that's a huge but, the weather is nice! Aside from a few rain storms every winter, it's clear.
It's the driving to Tahoe in the snow that's a challenge...
I’m interested in where driverless taxis will let passengers off. Parking is impossible. Generally, taxis just double park for the time it takes for passengers to exit. Will driverless cars do that?
Presumably they already are - the recent vote was to allow expansion of the service(s) rather than initiation of them.
I’ve seen a lot of Waymo cars some driving on their own and some with a person in the passenger seat in front, but never one with a passenger in back and no one in front. I assume you are right that it is happening but certainly not in large numbers. I’ve never seen a Waymo stopped letting a passenger off.
When I need to make a quick stop in a busy SF area and I’m with my wife, I or my wife will double park while the other hops out of the car. then the driver drives around and picks the other up when done. I wonder if Waymos can do something like that.
OK, I thought I was going crazy, so I spent some time tracking down Kevin's "longtime prediction."
In November, 2018, Kevin declared "Driverless Cars Are Here", based on Waymo's claim to be on the verge of a self-driving taxi service—which is to say, roughly the same evidence he's using today.
At the end of that post is a postscript: "To be more precise, I’m pulling in my prediction of when driverless cars will be available for purchase or lease by ordinary customers from 2025 to 2022." So we've already blown past Kevin's 5-yo prediction, and now he's moving the goalposts back.
In 2018, driverless cars for the masses were 4 years away. In 2023, they're 2 years away. Maybe by 2026, they'll be just 1 year away!
better than fusion energy
😉
Gravitationally-confined thermonuclear reactors work!
but they're hard to fit under the hood
An earlier "origin" might be https://www.motherjones.com/kevin-drum/2016/08/well-have-self-driving-cars-2025/
using "drum driverless 2025 site:motherjones.com" as the search parameters. Any number of other interesting results in that search. Including, but not limited to, https://www.motherjones.com/kevin-drum/2017/07/mass-unemployment-will-start-around-2025/ and https://www.motherjones.com/media/2013/05/robots-artificial-intelligence-jobs-automation/
Going a bit further back, and generalizing a bit, with "drum intelligence site:washingtonmonthly.com" yields: https://washingtonmonthly.com/2005/09/21/the-singularity/
In addition to Kevin's wildly wrong predictions, here is a fun list:
https://www.driverless-future.com/?page_id=384
There were a couple of more realistic predictions, but the overwhelming majority have already been missed by years.
I live in SF and I have been impressed and they have advanced further than I would have thought by now. I think you can say we've reached, as I understand the levels, level 4. I think that going from L4 to L5 will be harder than going from L0 to L4, though. Driving in limited areas in limited conditions is way different then me getting in my car in SF, tell it to drive me to NYC in December and I can sit back and sleep, read a book, etc. is way way off still.
I'm even beginning to question the business model because of that. It will take massive investment to get there and in the end, if it does actually work, it seems that fewer cars will be needed. Do the major automakers want to say, hey, we are doing all of this investment to sell fewer cars?
With that said, I'm only mildly against the action taken by the City Council in that it will just add more cars to the roads, similarly to what happened with the advent of Lyft and Uber.
I think humans will continue to have an edge over AVs but only in a very specific area. This area is not dealing with unusual situations on the road, per se. I think that will be handled.
No, it will be in dealing with specific unusual situations on the highway that are created by humans to have a specific and malicious effect. I can easily imagine humans engineering a way to hijack an AV by simply standing in front of the vehicle.
A human, once they have assessed the threat correctly, will switch modes and run the guy over who is standing in front of their car. For instance, when they see the guy coming from the side with a gun.
This is why I think that while AVs will come to have a strong role in interstate trucking, trucks will not be void of humans for long distances on the highway. It makes them too vulnerable to crazy stuff.
Ie, "The Johnny Cab Scenario" 🙂 https://www.youtube.com/watch?v=eAkeZqAN_qU
Consider what is lost. I live in a small city area with usually light traffic but crowded during tourist season. In a country that seems to be unable to cooperate on very much, and people self segregating into enclaves, it is reassuring to me how driving is the main exception in daily life where people act prudently, responsibly and (mostly) with courtesy. This dangerous, cooperative activity is one of the few things that shows human beings are capable of cooperation. This deadly dance, is arguably our most important social, and certainly our most time invested activity, and may show us at our best. And, without it our social distancing could increase exponentially.
road rage?
I think you haven't driven in suberban California recently.
I suspect the particularly hard part is the transition from driver-controlled to a mix of driver and driverless controlled to full driverless. One trope I've seen in sf is that drivers are allowed to be in charge, say, in strictly residential/local streets: trips to the local grocery store/pharmacy/dog groomer's/etc. But once you hit the feeder roads or motorways, that's it; you _must_ cede control to the grid. That being said, traffic planning is hard. And by hard, I mean computationally hard. Check out this video on why, paradoxically, adding more roads leads to more congestion, not less.
That link is a not-so-subtle promo, BTW. I'm always looking for good math content authored and explained by women.Don't get me wrong Grant Sanderson is a fantastic explainer and his production values are top tier. But a) you don't necessarily need all that razzle-dazzle video content to explain something well and in fact there are plenty of channels that get along just fine with the low-tech low-rez approach, and b) No offense, Grant, but you're a guy, and we need more gals out there to encourage women in the STEM (now STEAM) fields.
P.S. - Sorry for the abrupt right turn, but given my personal history, I've taken it as something as a personal mission to spread the word.
Check out Physics Girl. She's not active currently due to health issues, but I hope she recovers and returns. She's great.
Oh, yes, I'm already subcribed to her channel. Vi Hart is also really good and has just the sort of infectious enthusiasm that really pulls you -- or at least me -- in.
If you want something besides pure math, there's the Hasenpfeffer Lady a.k.a. Sabine Hossenfelder of Backreaction, who I can't recommend highly enough. She does both weekly science news and what she calls 'Science Without the Gobbledygook' which is more accurately translated as 'Science Without the Bullshit': She was an early, accurate, and above all, relentless critic of string theory bullshit (And drummed out of academia when it became apparent she was right all along. Running with the herd in the wrong direction is one thing. Pointing out the herd is running in the wrong direction being right is quite another.) Anyway, she's really good at cutting through the usual quantum weirdness woo with clear explanations of superposition, entanglement, many worlds vs hidden variables vs superdeterminism, etc.
_Without_ mentioning linear algebra. At all. 🙂
I second the recommendations for Physics Girl and Vi Hart, and add a recommendation for Rebecca Watson and finally, I'm afraid, vote against a recommendation for Sabine Hossenfelder. She's very good *when she sticks to physics* - but when she goes outside her field she's is sometimes atrocious. Witness her recent, ghastly video in which she attempted to explain trans issues.
Not knowing what STEAM was meant to include, my search for "stem vs steam" suggests that there is further expansion to "STREAM" Alphabetic expansion would seem to be an inevitable thing.
science, technology, engineering, mathematics
science, technology, engineering, arts, mathematics (steampunk?)
science, technology, reading, engineering, arts, mathematics
I'm behind the times, apparently. Thanks for the update.
STREAM --it sounds like --education doesn't it?
Ultimately, we end-up with a song by Big Bird: https://www.youtube.com/watch?v=qTvhKZHAP8U
Can't wait for the inevitable moment when a hacker from a foreign or domestic terrorist group - or even just a bored/overworked employee - puts errant/nefarious code into the traffic grid's central computer that all these cars are going to be relying on, and then watching all the automated cars suddenly either crash into each other or stop working completely.
There's a reason why you always need humans somewhere in the loop and able to stop what the machines are doing. Have none of these automated driving fan boys ever seen Terminator or WarGames or basically any tech movie set in the future? Letting the computer control it all almost never works out in those movies.
"I think humans will continue to have an edge over AVs but only in a very specific area. This area is not dealing with unusual situations on the road, per se. I think that will be handled."
Translation: Truly 100% driveless automobiles won't happen until you can equip them with general near-human equivalent AI with domain-specific knowledge. [1]
[1] At which point, I think I'd prefer an R. Daneel / Giskard valet to chauffer me around. [2]
[2] Will there ever be such a thing as an R. Jeeves to save well-meaning idiots from themselves? And R. Spodes to keep not-so-well meaning idiots in line?
A debate for the ages: Is Kevin more credulous about self-driving cars, or about AI?
A short update from a Tesla owner who has had the FSD feature since 2019, and was a beta driver on the full version which is now generally available to all.
In driving, simply, humans are probably at, if we have to put a percentage on it, 99% effective. At least, maybe really 99 point something.
I haven't ridden in a Waymo, but Waymo and Cruise use geo-fenced hyper accurate gps maps of streets to assist, whereas Tesla uses maps only to identify where, for example an intersection generally is -- i.e., like humans use maps. Tesla uses sensors to actually drive the car.
So, on our "percentage" based analysis, Tesla is easily 90% there, but 90% is nowhere near good enough to do entire trips without a driver.
It's somewhat disappointing to me, through a combination of Musk being an emotional 13 year old in many ways and having no sense of marketing at all, that the incredible advances in Tesla's FSD software suite is not hearlded as what it is, one of the most outstanding achievements ever. The current version at 90% can get the seven hour trip from LA to Lake Tahoe where the human only has to driver for a period of time meausured in minutes. But those minutes matter.
But as many have said, that last 10% is going to be tough. That is for the simple reason that the current Tesla sensor suite reacts far far better than a human now, but the key question is "reacts to what?" In order to react to 99% of all situations you have to write code for 99% of all situations.
While we are at it, this isn't "AI" in any sense - none of the systems use experience to anticipate an action. The sensors just react. Its interesting in that from what I have seen reaction is enough, as long as you have 8 cameras and other sensors.
According to a stat above, US drivers drove 3.2T miles in 2022. 99% isn't good enough: that implies that 32B miles will still require humans to intervene to avoid some kind of failure of the automated system. That's a lot of effing miles to fall outside "full self-driving" capabilities, IMO.
The last time I checked, the successful model was to have exquisitely detailed and frequently updated mapping. Which is to say, exquisitely expensive mapping. Which is to say that this will only happen in areas that are both densely populated and affluent. (Should we add "White" to this list? Probably.) This in turn is to say that the model is very much not scalable.
Are they using a different technology that can be generalized? Inquiring minds want to know.
You can do without the mapping if the car has enough sensors and cameras to Make It Up As It Goes Along, but you need the roads, markings, signage and signals to be of Top Engineering Quality... which for many poorer parts of the country are all lacking. (Really poor municipalities and counties often lack any markings on the roads, and feature lots of traffic lights and signs out of order or missing/damaged. But I guess in theory the people who will be able to afford fully automated cars won't be going to those places anyway?)
My town just had a pretty impressive storm system pass through. I'm not sure, but it might have made the national news. If you saw a photograph showing a line of snapped utility poles lying gracefully across a major roadway, that was about a mile from my house. Less dramatic are the traffic lines pushed just a bit out of alignment, so they don't quite line up with the roads they are controlling. This is not that big a deal for a human, but I can easily picture a driverless car obeying the wrong light. This will get tidied up soon enough. We are not a poor jurisdiction. But this is not at the top of the to-do list. I give it about two weeks.
The cars will be good enough for relentlessly suburban Orange County, where nobody walks or bikes and the infrastructure is all well-maintained and the weather generally is no snow and no rain every day, which is good enough for Kevin to declare Mission Accomplished.
Everyone else who lives someplace that isn't similar to Orange County will experience more frustration than contentment with their automated cars, but That's None Of Kevin's Concern.
I was watching local news last night and the pro faction for autonomous vehicles included bicyclists. I was surprised. But apparently there is more to fear from drivers.
1995: "Speech recognition software is about 92% of the way there."
2023 YouTube caption: "well with delta magnatar clothesline fish ridge just off the elf hole here she won't spell any beer"
Fully self-driving cars will enable a guy to go to a bar, get totally hammered, and then drive home alone and safely. The key is that the car needs to be able to understand drunken, slurred speech. I hope that they are working on that critically important detail.
Car waits for the guy to pass out then decides if he needs to go to the ER or just park in his driveway.
You do realize that Waymo has detail mapped those places within an inch of their lives?
"All that's left is non-street driving: parking lots, drive throughs, dirt roads, and so forth."
Nighttime, rain, snow, overcast (did you ever wonder why they test only in the desert Southwest? It isn't just because they live in the Valley.), free of traffic barriers, pedestrians, and policemen . . .
Kevin: "Look how fast we solved all the easy parts! Surely that means we can knock the hard parts out over the weekend!"
Yep, they are here.
https://www.ktvu.com/news/driverless-cruise-cars-cause-traffic-jam-day-after-mass-expansion-approved-in-san-francisco