After 50 million miles, Waymos crash a lot less than human drivers
- Thread starter JournalBot
- Start date
That's how Waymo works, iirc. They don't use only their sensor suite; they're also driving to a very detailed 3D map that includes curbs, poles, trees, hydrants, drains, lane lines, buildings, etc. This map constantly updates as fleet sensors detect changes in the environment (eg, construction or new stripes). So in principle, the AV can drive on a freshly resurfaced street (or snow-packed street) and stay to the lanes even without seeing the stripes - as long as it can see the curbs and light poles, etc. This is a big reason why they're so tightly geo-fenced: they need to build that ultra-hi-res map before serving an area.
Iirc, the main issue with driving in heavy snow or rain, is that precipitation interferes with lidar returns, causing confusing for the object classification algorithms. To a degree, that's a software problem, and has improved enormously in the last several years. But I think there's probably a performance plateau for which the best answer is "don't drive now" - just like there is for a real person with eyeballs.
Upvote
15
(15
/
0)
I have yet to take one, but I see them all over my neighborhood in West Los Angeles, and would love to see the data about how they handle our incredible lack of left turn signals. Here a yellow light means "move into the intersection and pray opposing traffic stops so I can turn left before cross traffic starts". My sister doesn't like to visit me because she finds it too stressful to drive here, which is why I could see a Waymo sitting for multiple lights unable to turn left.
Upvote
6
(6
/
0)
I mean if you actually finished reading the article, you would have known some of the effort taken to figure try to make the miles comparable (there were links for further information). In general, most of the inaccuracies are favorable to humans (i.e. it makes their accident numbers look better). For instance the article mentions that the human miles include highway milesyou can drive a lot more miles on a highway since you mostly just staying in the same lane at the same speed and opposite direction of traffic is separate and can't turn in front of you compared to waymo's constant navigation of city traffic and intersections).
They just took all human drivers in the same area and time period. I'm sure you believe yourself to be above average, but bad drivers can still crash into you and I will claim that Uber/Lyft drivers tend to be below average drivers (I don't have data on this, but it "feels" like it to me).
"That's the news from Lake Wobegon, where all the women are strong, all the men are good-looking, and all the children are above average."
Upvote
17
(17
/
0)
I have little doubt that Waymo drivers are safer than the average human in current markets. However, from what I recall, these tests have been limited to regions in which Google/Waymo created absurdly detailed maps of the driving environment. How will that translate to areas that have not been mapped by specialized mapping vehicles with millions of dollars in sensors and a team of engineers to review and fix map edge cases?
Upvote
-16
(4
/
-20)
I mean you can. I took a couple of rides in a waymo vehicle in San Francisco yesterday with 4 passengers (the max). Both times we all exited the passenger side, and this was full size American adults. Back in the 90s, minivans usually had only one rear door on the passenger side. This was specifically to keep kids exiting curb side. That would be restrictive for cities with one way streets where you might pull over on the left
Upvote
5
(5
/
0)
To replace all humans with Waymo maybe... but what about to replace the lower 25-percentile of driver. Also remember that you can get into an accident because someone else who is a bad driver runs a red light or rear ends you. Waymo posted a macabre study looking at all the fatal multi-vehicle accidents in Phoenix. They looks at what-if the human driver was replaced by a Waymo driver, not surprisingly when the "initiator" vehicle was a Waymo 100% of accident were avoided by Waymo don't speed, get tired/intoxicated/distracted.
The more interesting thing, is that if Waymo was in the "responder" vehicle, 82% of the fatal accident were also completely avoided because Waymo can see 300m in all directions all of the time and notice the incoming speeding car and avoid it. In an addition 10% the Waymo was able to reduce the severity to sub-fatal levels.
Upvote
17
(17
/
0)
"At fault" figures for accidents is a poor standard for measurement. A key to driving is being able to adapt and avoid an accident when someone else on the road screws up. Drivers do make mistakes (myself included), and it is important to be able to avoid those accidents even if it is the other's fault. Accordingly, I'd stick purely to the number of accidents per driven mile, regardless of fault.
Upvote
-17
(5
/
-22)
People have mentioned that they would prefer good public transportation, but why couldn't this be the public transportation?
On routes that lots of people travel, trains, buses, and subways make a lot of sense. But out in the suburbs, I've noticed that on all but a select handful of routes, buses are usually driving empty or near empty. Taking a vehicle straight to your destination has to be a lot more efficient than having a huge 10mpg bus drive around near-empty all day. It is counter-intuitive to think of buses as being inefficient, and polluting, but in this particular case I think they are.
Back in the 1990s, the county here (Santa Clara, Calif) realized this and tried a 'call to drive' system where instead of having buses drive on the less crowded routes every day, you'd call when you needed transportation and the bus would come pick you up. It didn't catch on so they cancelled it, but now with the Internet and driverless cars, they could do this again with a deal with Waymo, and they'd probably end up saving money and polluting and using fuel a lot less.
Hey Santa Clara County--look into making a deal with Waymo! This is where Waymo is headquartered and did their initial testing many years ago, so it should be pretty easy to set up.
On routes that lots of people travel, trains, buses, and subways make a lot of sense. But out in the suburbs, I've noticed that on all but a select handful of routes, buses are usually driving empty or near empty. Taking a vehicle straight to your destination has to be a lot more efficient than having a huge 10mpg bus drive around near-empty all day. It is counter-intuitive to think of buses as being inefficient, and polluting, but in this particular case I think they are.
Back in the 1990s, the county here (Santa Clara, Calif) realized this and tried a 'call to drive' system where instead of having buses drive on the less crowded routes every day, you'd call when you needed transportation and the bus would come pick you up. It didn't catch on so they cancelled it, but now with the Internet and driverless cars, they could do this again with a deal with Waymo, and they'd probably end up saving money and polluting and using fuel a lot less.
Hey Santa Clara County--look into making a deal with Waymo! This is where Waymo is headquartered and did their initial testing many years ago, so it should be pretty easy to set up.
Upvote
8
(9
/
-1)
Pesky humans. Always running into our robotic overlords. Don't make them angry.
Upvote
3
(3
/
0)
No shit, there I was...
Back in the '90s four of us went on a business trip and rented a minivan. We had it for 2 days before we noticed it also had a driver side sliding door.
Upvote
2
(3
/
-1)
You mean this data?
Seems like they release it quite regularly. Per that same webpage:
Feel free to back up these assertions with evidence.
Provide some proof that what you’re saying is what Waymo is doing.
Did you even read the article you linked to?
Nowhere in the article does it say that the cars were creating a hazardous situation.
And they are selecting comparable miles.
Has this actually happened or are you just clutching at pearls of your own design?
Upvote
20
(22
/
-2)
Not sure why the down votes, unless it is due to departing from WAYMO. It's a trivial exercise to find a Musk cult member insisting that Tesla is actually fas ahead of WAYMO and nearly SAE level 5 while WAYMO is really only Level 2. Musk claims that Tesla's AP and FSD are 8 times safer than manual drivers, but true to form, he refuses to show any actual evidence, and he desperately wants to be excused from the Federal crash reporting requirement. The latter may happen, since he has the (fake) governmental position of being able to kill the funding for the NHTSA.
https://www.nhtsa.gov/
Last edited:
Upvote
-1
(2
/
-3)
Even ignoring the cost issue, the safety issue differences are rather large. I once rode in an Uber (NEVER again) who continually over sped and routinely tailgated every vehicle ahead. I was extremely relieved when arriving at my destination without a crash, and zero tip.
Upvote
2
(2
/
0)
TLDR: Are people actually interested in finding out but just too lazy to read the article and click on the links for more data. Or are they Just Asking Questions?
Doherty, S. T., Andrey, J. C., & MacGregor, C. (1998). The influence of passengers, time of day and day of week on accident rates. Accident Analysis & Prevention, 30(1), 45-52.
Edwards, J. B. (1998). The relationship between road accident severity and recorded weather. Journal of Safety Research, 29(4), 249-262
What they did was exclude the freeway miles for Waymo (because they have safety drivers) but included highway miles for humans (because it hard to pull out) but that actually helps the human accident per vehicle miles travelled by a lot because you are a lot less likely to get into an accident per mile on a highway
The linked article linked to the Waymo study were they do break things out by time of day (I can try and find the link if you are really interested, or you could look it up yourself), they find humans are a lot more likely to get into accidents at night (tired/fatigue/partying/phone they are not sure). If you look at the citations you can see they made effort to control for these factors
Doherty, S. T., Andrey, J. C., & MacGregor, C. (1998). The influence of passengers, time of day and day of week on accident rates. Accident Analysis & Prevention, 30(1), 45-52.
Edwards, J. B. (1998). The relationship between road accident severity and recorded weather. Journal of Safety Research, 29(4), 249-262
Upvote
17
(18
/
-1)
The question Waymo is trying to answer with these data releases is, should Waymo be allowed to operate in X city? Comparing Waymo to average driver performance seems appropriate in that context. I think an even better analysis would be to look at the type of driver whose miles will be replaced by Waymo: this obviously is heavily biased towards for-hire drivers: Ubers, cabs, car services, etc.
So rather than cherry pick “good” drivers, I’d want to compare Waymo performance to for-hire drivers. But of course Waymo is unlikely to cut the data this way, even if they still win, because it puts a spotlight on their actual business case: replace human for-hire drivers with robots. Which is another reason a city might decide not to allow Waymo to operate.
Upvote
8
(9
/
-1)
You could have stopped right there.
Conditions like obeying ALL the traffic laws instead of treating them as optional and annoying?
Upvote
12
(14
/
-2)
This was literally covered in the second half article with link to more data for you Just Asking Question people.
Upvote
12
(13
/
-1)
Do you also worry about what will happen to the worldwide monetary system if the tooth fairy is discovered to be real?
Upvote
6
(9
/
-3)
Good point, ambulances can often travel at higher speeds than other slower traffic! /S
Upvote
2
(4
/
-2)
Upvote
11
(13
/
-2)
It was rather trivial to find this link:
https://waymo.com/blog/2023/07/past...n-drivers-speed-in-san-francisco-and-phoenix/
Upvote
4
(5
/
-1)
Don’t selectively edit what the OP said.
They framed their question appropriately and I think that their concern about a hostile state or individual compromising the cybersecurity integrity of a Waymo vehicle or its entire fleet is legitimate.
Upvote
-1
(7
/
-8)
Upvote
1
(1
/
0)
If there were transponders in cars, and along the roads, how could that increase safety in self driving vehicles? I could also see pedestrians and bicyclists carrying transponders to increase their safety.
Upvote
-12
(1
/
-13)
To be really useful, this can't be random.
We should be looking at the same circumstances. How many miles per accident on the streets that Waymo is driving. If there are more human accidents on the ice-covered hills of Illinois (where Waymo doesn't drive) than on the wide-flat streets of Phoenix (where it does), then you are getting misleading results.
The numbers may come out even better for Waymo, or worse, or the same. I don't know. But I do know that Waymo's difference in where they run from "randomly in the US" means it's not an ideal comparison.
Upvote
-7
(2
/
-9)
It shows what it will be. The geofences are expanding (in both current locations and adding new ones). At some point the various geofences will merge. The miles between intervention is improving. Costs are coming way down.
Upvote
8
(8
/
0)
Make some reasonable estimates of each variable and then do some math and you'll have an answer.
One of the few times in your life where algebra 1 or whatever pays off.
Last edited:
Upvote
1
(3
/
-2)
From what I read, it was “idiot human drivers crash way more than Waymo”. You’d probably need a few years data to identify good drivers, and I’d be curious how waymo compares to good drivers.
Upvote
-6
(1
/
-7)
Since I did not selectively or otherwise edit what the OP stated, but rather quoted him/her to demonstrate how paranoid some people can be: But what if the lizard people taking over the world governments are real? For example, what would happen with human drivers if every traffic light in the world suddenly turned permanently green? What if the moon is farther away than the sun, as some religionists claim, wouldn't that show that NASA and moon landings are frauds? There are no lower limits for self-induced paranoia. Thank you for showing where you stand on this very important and pressing issue. JAQing off is a common hobby for obscurantists.
Upvote
6
(8
/
-2)
If you are caught, you won’t be done just for speeding. For example waymo will have a license which may be invalid if you override the speed limit. And suddenly you drive a car that is not declared roadworthy, has no valid insurance, and so on. So this will be very expensive.
Upvote
0
(1
/
-1)
IT won't "translate" at all, because those detailed maps are integral to the driving task, at least right now.
I don't know Waymo's long-term strategy, but the reason they don't currently deploy just anywhere at anytime is because they first collect tons of map data. (I don't know if that requires special mapping cars nowadays, but that mapping used to be done mainly with the sensor suite on regular AVs with drivers.) The collation processing is mostly automated, obviously, with engineers monitoring it for quality and to sort out your edge cases. After deployment, the map is updated as the fleet drives through it, using their own sensor data compared to the map. The AVs flag both temporary changes that might impact fleet behavior (eg, nearby construction sites where workers or heavy equipment is likely to be encountered); and permanent changes that eventually lead to updates in the map itself (eg, re-striping, changes in traffic lights, new curb cutouts, etc).
Upvote
5
(5
/
0)
TLDR: It's good to "Be Curious, Not Judgmental".
Waymo gives the direction/speed of the vehicles a few second before and during impact. To me it doesn't look like that worry is substantiated in the data. Waymo gives sensor/video data to CA DMV, so I don't think they are going to try an lie and be contradict like a police officer with their own body camera, or hide the video like what Cruise did (as that coverup basically ended Cruise as a company).
AFAICT, that was really only true of Waymo test vehicle from years ago, basically before they started the robotaxi service to the public and it's been 5 years since then. The description of rear ends even in this paper using data from 2019 didn't seem to reflect the scenarios you are suggesting
If you went digging, you can find that information.
Waymo gives the direction/speed of the vehicles a few second before and during impact. To me it doesn't look like that worry is substantiated in the data. Waymo gives sensor/video data to CA DMV, so I don't think they are going to try an lie and be contradict like a police officer with their own body camera, or hide the video like what Cruise did (as that coverup basically ended Cruise as a company).
As the article mentions, 50M miles is 70 lifetimes of driving on local city road and there were only 60 accidents and so it is getting less than 1 accident a human lifetime.
Not really, it might shift some of the liability to other types of accidents, but if they are less accident over all the financial/legal liability is being reduced. AFAICT, the one case were the Waymo might do something weird and cause a human driver behind them to rear-end up (which Waymo isn't liability for but is still more accidents) doesn't seem to be happening.
Upvote
12
(13
/
-1)
Upvote
0
(1
/
-1)
As I mentioned in another post above, Waymo actually did macabre study on what happens" when a drunk driver runs a red light" type scenarios. They took all the fatal accidents in their Phoenix operation area over 10 years and simulated what would happen if Waymo was there. Not surprisingly if Waymo replaced the "initiator" vehicle 100% of accident were completely avoided. More interesting, if they replace the "responder" vehicle, 82% of the accidents were also completely avoided as the Waymo can pay attention in 360 degree constantly and see 300m away. 10% more of the crash were reduced to non-fatal levels.
Sorry to be repetitive, but I am seeing a lot of the people aren't even reading the entire article before commenting so I am not assuming they will see posts that don't directly respond to them.
Last edited:
Upvote
11
(11
/
0)
There may be changes to insurance premiums where bad drivers get substantial financial pressure to get a self driving car. (In Germany there is enormous pressure to have liability insurance for substantial amounts, at least a million. It’s basically impossible to drive without insurance except with a stolen car. That seems different in the USA. )
Upvote
1
(2
/
-1)
They could try that. However, getting the cars to work there is probably far beyond current technology and even if they did work, the riders would still be living in Minneapolis (that's the American word for Winnipeg) which is... not great.
It might be that the solution is to move all non-essential personnel to warm weather cities where the people could enjoy a better quality of life including self-driving vehicles. Nowadays a lot of stuff can be done remotely, so there isn't a huge benefit of having lawyers, accountants, salespeople, and many others exposed to frigid conditions.
Another option, I suppose, would be to warm the climate and extend tropical conditions northward to Minneapolis. Minneapolis is 1,110 km north of Memphis, the closest place with a decent climate, so this might be possible.
Upvote
-2
(2
/
-4)
I’d leave it all to the insurance company. Let them tell you “$x a month if you drive yourself, $y a month if the car drives”. They have the statistics. They charge based on averages, and they will figure out the averages after a while.
Upvote
2
(2
/
0)
Having been in more than one South American city, even today's rotten drivers have trouble competing with the cab drivers in many SA cities. It was exceptionally routine for the SA cab drivers to honk their horn, religiously cross themselves, and blast through every single stop sign, no exceptions. Even looking for cross traffic never happened. At traffic lights they would stop but lane split like it was required by law. 4 lane roads became at least 5 lanes, sidewalks (if they existed) became traffic lanes, etc.
Upvote
2
(2
/
0)