I'm an ATLien who used to work in 911 and still have a lot of friends that do. About a month ago one of my friends sent me a picture of a Waymo that managed to wander its way in to the Grady (the...
I'm an ATLien who used to work in 911 and still have a lot of friends that do. About a month ago one of my friends sent me a picture of a Waymo that managed to wander its way in to the Grady (the city's major level 1 trauma center) ambulance bay, and then just stopped because it wasn't sure what to do, blocking some spaces and also blocking a few trucks from being able to get out. Apparently it took them almost half an hour to get it out of the way.
These sort of incidents (both my described one, and the article) don't exactly inspire confidence that this technology is ready to be out in the public, for me.
I feel bad for shilling on Google's behalf (a company I mostly try to avoid where I can) - but consider that fewer people are arriving at the emergency room because of self-driving cars.
I feel bad for shilling on Google's behalf (a company I mostly try to avoid where I can) - but consider that fewer people are arriving at the emergency room because of self-driving cars.
This may be true in California, Nevada, Arizona but sudden downpours and flooded roads are common in many states. Failure to recognize a flooded road is a problem that needs to be solved before...
This may be true in California, Nevada, Arizona but sudden downpours and flooded roads are common in many states. Failure to recognize a flooded road is a problem that needs to be solved before roll out.
Honestly... Yeah, I mostly agree, but on the other hand, there are two key points here that may tip the scales. One, from every statistic I've seen, waymos are safer drivers than people by a...
Honestly... Yeah, I mostly agree, but on the other hand, there are two key points here that may tip the scales. One, from every statistic I've seen, waymos are safer drivers than people by a decent margin. Two, it's virtually impossible to test and design for every potential scenario in the real world. You can get incrementally closer, but as you approach perfect, the time and cost involved goes up exponentially, approaching infinity.
Usually, what happens in these situations is that the problems happen, the issue is worked around (like suspending service), and the problem is then fixed fleet wide. Then the next thing pops up.
In the mean time, people taking safer robo taxis means fewer people are taking less safe human driven cars.
Obviously it's within Googles best interest to get these things out as fast as possible and starting to make money, but we as a society (represented by regulators) also need to balance the cost of slightly buggier self driving cars being in the road with the cost of more way buggier human driven cars being on the road.
Waiting until they're perfect isn't an optimal solution, because that would mean no self driving cars ever.
I agree with you generally but also maintain that testing should have found and prevented the flooded road bug before rolling out the cars in stormy territory. Those numbers you quote might well...
I agree with you generally but also maintain that testing should have found and prevented the flooded road bug before rolling out the cars in stormy territory.
Those numbers you quote might well compare humans driving in snow vs robots driving in the desert.
Aren't they still in a testing phase? They haven't even be available for a year in Atlanta, so this all seems normal and expected. Unless I missed something, nobody has been harmed nor directly...
Aren't they still in a testing phase? They haven't even be available for a year in Atlanta, so this all seems normal and expected.
Unless I missed something, nobody has been harmed nor directly endangered. Waymo voluntarily suspends service during severe weather.
It's a very good thing that this flaw was discovered. Flash flooding due to geographically distant rainstorms is a safety issue in the desert Southwest where the cars are well beyond the...
It's a very good thing that this flaw was discovered. Flash flooding due to geographically distant rainstorms is a safety issue in the desert Southwest where the cars are well beyond the introductory testing phase.
There is a big difference between "waiting until they're perfect" and "waiting until they're safe enough encountering known common hazards in the area where they're being rolled out", though, and...
There is a big difference between "waiting until they're perfect" and "waiting until they're safe enough encountering known common hazards in the area where they're being rolled out", though, and failing to meet the latter threshold before rolling them out commercially is imo negligence and even at best simply isn't making the roads safer. To my knowledge they haven't been rolled out anywhere where there's heavy snow in the winter, and it would be ludicrously evil to roll them out somewhere like that before you had a certain ability to handle that common weather condition with a certain degree of safety.
Yes, training them to handle these circumstances is a difficult task, but that difficulty doesn't absolve the companies from ensuring that their vehicles are operating safely where they've been rolled out. This isn't just a matter of "slightly buggier cars" -- being unable to handle common severe weather conditions in an area endangers customers and bystanders. If your goal is to actually increase safety by replacing fallible human drivers, you can't ignore huge safety defects in self-driving cars. The fact that this is the case is why they're backing out of these locations now until they can find solutions to flooded roads that are safe enough, and they should have had better testing in extreme conditions like this to recognize this problem earlier, before rolling out the paid taxi service to consumers.
This is a "pause" while severe weather is likely. The comparison is between driving in such conditions or not, rather than whether they will be in Atlanta or not. It seems similar to how airlines...
This is a "pause" while severe weather is likely. The comparison is between driving in such conditions or not, rather than whether they will be in Atlanta or not. It seems similar to how airlines cancel flights when the weather looks too bad?
They are relying on weather forecasts to avoid driving in severe weather, but it turns out they needed to be more conservative about that, as well as improving their cars' ability to cope when the weather suddenly turns bad.
Yes, as I see it every trip taken in a Waymo reduces the odds of an injury. I’ve also seen them force nearby human drivers to be safer. They drive slowly and can see around corners better than...
Yes, as I see it every trip taken in a Waymo reduces the odds of an injury. I’ve also seen them force nearby human drivers to be safer. They drive slowly and can see around corners better than humans. A Waymo showing down near an alley or intersection lets drivers know something is approaching from out of sight.
After the grand-parent comment on blocking ambulance bay, I almost mis-read your comment to mean we have the same number of injuries, but fewer of them survive long enough to arrive at the ER...
After the grand-parent comment on blocking ambulance bay, I almost mis-read your comment to mean we have the same number of injuries, but fewer of them survive long enough to arrive at the ER because of self driving cars.
But you meant to say that self driving cars are much safer than human driven cars.
Ignorable aside to explain personal bias: I do wonder if this is because we're having fewer accidents per million cars on the road? Or that accidents are less fatal over same period or over same number accidents? Or that we have fewer cars and hence fewer accidents? Can these factors be attributed to anything else such as high gas prices or people working from home or drinking/partying/driving drunk less? The onus is on me to look up this information and convince myself again and again every time this comes up, so please feel free to ignore.
I will admit my bias is anti big tech and I'm very skeptical of how quickly these things have been approved to be on the road with humans. Eg, we still don't even have women crash test dummies after decades but this we approved this fast? I'm sure that big tech will release all kinds of lies, damn lies and statistics skewed to show them in a good light.
I can only imagine how robotaxis will behave once deployed to a place with chronic snow 30% of the year. Seems like our roads have no end of edge cases for machine vision with no underlying...
I can only imagine how robotaxis will behave once deployed to a place with chronic snow 30% of the year. Seems like our roads have no end of edge cases for machine vision with no underlying understanding of the world. Of course human drivers are imperfect, but it's interesting to consider that most humans won't make this mistake as long as they realize that water depth is very difficult to judge. Whereas the robotaxis seem to just... not perceive the water? I wonder why LIDAR doesn't tip them off.
Bodies of water absorb LiDAR, so no light back to the sensor and it will just report a void there. Sometimes you'll get the odd return, particularly on the ground under the water for shallow...
Bodies of water absorb LiDAR, so no light back to the sensor and it will just report a void there. Sometimes you'll get the odd return, particularly on the ground under the water for shallow water. I see it pretty often in my daily work. You'll need something else to sense it.
How often during regular trips do these robotaxis get a lidar void and still drive ahead? I’m suddenly imagining a cartoon style driving off a cliff, pause in midair, and then plummeting with that...
How often during regular trips do these robotaxis get a lidar void and still drive ahead? I’m suddenly imagining a cartoon style driving off a cliff, pause in midair, and then plummeting with that “oh no not again” look on the cartoon face
Stopping erroneously is, while much safer than driving into a flood, not ideal. And the cars can't stop unconditionally for puddles, or any schmuck with a garden hose can DoS them. (Furthermore:...
Stopping erroneously is, while much safer than driving into a flood, not ideal. And the cars can't stop unconditionally for puddles, or any schmuck with a garden hose can DoS them. (Furthermore: and make it look like an accident.)
To be fair here, "don't drive into floods" is unironically challenging for human drivers, as well. It's tough to judge the depth of a body of water, so a great deal of the "is this a dangerous flood or a shallow puddle" decision comes down to context and heuristics. What's your extrapolated topography of the street? If the water is flowing, what does it look like? Have any other cars gone through it? Is there a schmuck with a garden hose guffawing next to the road? Etc.
There is definitely a reason these trials have heretofore been in socal or the southwest, though.
Waymo maps the roads where they will be driving. I imagine they could predict where flooding might happen based on things like elevation to bridge underpasses, erc. Also, the cars could compare...
Waymo maps the roads where they will be driving. I imagine they could predict where flooding might happen based on things like elevation to bridge underpasses, erc. Also, the cars could compare current conditions to what the road looks like when it's dry.
Ah our work is almost the total opposite of a robotaxi, we collect it from a flight and process that. Not real-time or solely machine analyzed. Void would be wherever there is open air, so going...
Ah our work is almost the total opposite of a robotaxi, we collect it from a flight and process that. Not real-time or solely machine analyzed. Void would be wherever there is open air, so going towards flooding, it would look like a cliff in lidar. Even under less severe flooding, it sounds like they're still panicking after "driving off the cliff". I don't think it'd happen often during normal trips, probably why Waymo didn't already have a solution prepared (other than the total lack of regard for safety).
If you would describe water to an alien who has never seen it, would they be able to identify it? It’s extremely difficult to identify by computers. Some types of sensors might perceive floodwater...
If you would describe water to an alien who has never seen it, would they be able to identify it? It’s extremely difficult to identify by computers. Some types of sensors might perceive floodwater as solid ground.
It’s this kind of thing which is the reason why I was against the concept of building out self driving cars that work like this instead of on closed purpose-made roads. Computers are really bad at identifying hazards and making complex decisions. It’s practically a miracle that Waymo has a self driving platform that operates as good as it does right now.
I think that discredits the work of the engineers. At some point you have to admit the injury rate is low because they've done an amazing job and has nothing to do with luck. It seems that, for...
It’s practically a miracle that Waymo has a self driving platform that operates as good as it does right now.
I think that discredits the work of the engineers. At some point you have to admit the injury rate is low because they've done an amazing job and has nothing to do with luck. It seems that, for now, they should stick to places with famously easy weather (SF, LA, Phoenix). But they'll figure it out eventually. I've been on team self-driving for decades now. Humans have only been driving because we haven't had a better alternative. I look forward to a future where someone driving a car is viewed as incredibly selfish.
Something can be a miracle of engineering as well! I look forward to a day when we don’t have absurd amounts of private cars with human drivers in them.
Something can be a miracle of engineering as well!
I look forward to a day when we don’t have absurd amounts of private cars with human drivers in them.
This is one of the touted uses of Google's "world model", Genie. They're using it to teach their cars about unlikely conditions: large animals on the road, conflagrations, tornadoes, and indeed...
Seems like our roads have no end of edge cases for machine vision with no underlying understanding of the world.
This is one of the touted uses of Google's "world model", Genie. They're using it to teach their cars about unlikely conditions: large animals on the road, conflagrations, tornadoes, and indeed flooding.
Maybe LIDAR has weird reflections off water? Whatever it is, I imagine they will take a few weeks or months to fix this and it will stay fixed. It doesn’t seem like it would be as difficult as...
Maybe LIDAR has weird reflections off water? Whatever it is, I imagine they will take a few weeks or months to fix this and it will stay fixed. It doesn’t seem like it would be as difficult as kangaroos.
The one benefit I can see to this is it might place the correct stigma on driving during bad weather. While humans can probably navigate heavy snow or rain compared to a "robotaxi", should we?
The one benefit I can see to this is it might place the correct stigma on driving during bad weather. While humans can probably navigate heavy snow or rain compared to a "robotaxi", should we?
Sometimes you have no choice -- for instance, if I wouldn't drive in snow, I would literally never leave my house some weeks! And plenty of in-person jobs basically require employees to take on...
Sometimes you have no choice -- for instance, if I wouldn't drive in snow, I would literally never leave my house some weeks! And plenty of in-person jobs basically require employees to take on that risk (some fairly, like firefighters, some less so, like in-person office gigs). But I do generally agree that we underestimate the risk!
Also, there is something to be said about prepardness of the region. An inch of snow is catastrophic to Texas, but in the northeast there might be a 2 hour delay if the schools are feeling generous.
Also, there is something to be said about prepardness of the region.
An inch of snow is catastrophic to Texas, but in the northeast there might be a 2 hour delay if the schools are feeling generous.
Waymo also halted service in Dallas and Houston because of severe weather across Texas this week, the company confirmed to TechCrunch late Thursday. The expansion was first reported by Bloomberg News.
One of Waymo’s robotaxis was spotted driving through a flooded street in Atlanta, Georgia on Wednesday before it ultimately got stuck for about an hour, according to local news reports. The vehicle was recovered and removed from the scene, Waymo told TechCrunch. Waymo says it paused service in the city, just like it has in San Antonio, Texas, while it figures out a solution.
[...]
But even those precautions apparently were not enough to stop the Waymo robotaxi from entering the flooded intersection in Atlanta. Waymo told TechCrunch on Thursday that the storm in Atlanta produced so much rainfall that flooding was happening before the National Weather Service had issued a flash flood warning, watch, or advisory. The company said its fleet those alerts are part of a larger set of signals it relies on to prepare the vehicles for poor weather.
I'm an ATLien who used to work in 911 and still have a lot of friends that do. About a month ago one of my friends sent me a picture of a Waymo that managed to wander its way in to the Grady (the city's major level 1 trauma center) ambulance bay, and then just stopped because it wasn't sure what to do, blocking some spaces and also blocking a few trucks from being able to get out. Apparently it took them almost half an hour to get it out of the way.
These sort of incidents (both my described one, and the article) don't exactly inspire confidence that this technology is ready to be out in the public, for me.
I feel bad for shilling on Google's behalf (a company I mostly try to avoid where I can) - but consider that fewer people are arriving at the emergency room because of self-driving cars.
This may be true in California, Nevada, Arizona but sudden downpours and flooded roads are common in many states. Failure to recognize a flooded road is a problem that needs to be solved before roll out.
Honestly... Yeah, I mostly agree, but on the other hand, there are two key points here that may tip the scales. One, from every statistic I've seen, waymos are safer drivers than people by a decent margin. Two, it's virtually impossible to test and design for every potential scenario in the real world. You can get incrementally closer, but as you approach perfect, the time and cost involved goes up exponentially, approaching infinity.
Usually, what happens in these situations is that the problems happen, the issue is worked around (like suspending service), and the problem is then fixed fleet wide. Then the next thing pops up.
In the mean time, people taking safer robo taxis means fewer people are taking less safe human driven cars.
Obviously it's within Googles best interest to get these things out as fast as possible and starting to make money, but we as a society (represented by regulators) also need to balance the cost of slightly buggier self driving cars being in the road with the cost of more way buggier human driven cars being on the road.
Waiting until they're perfect isn't an optimal solution, because that would mean no self driving cars ever.
I agree with you generally but also maintain that testing should have found and prevented the flooded road bug before rolling out the cars in stormy territory.
Those numbers you quote might well compare humans driving in snow vs robots driving in the desert.
Aren't they still in a testing phase? They haven't even be available for a year in Atlanta, so this all seems normal and expected.
Unless I missed something, nobody has been harmed nor directly endangered. Waymo voluntarily suspends service during severe weather.
It's a very good thing that this flaw was discovered. Flash flooding due to geographically distant rainstorms is a safety issue in the desert Southwest where the cars are well beyond the introductory testing phase.
There is a big difference between "waiting until they're perfect" and "waiting until they're safe enough encountering known common hazards in the area where they're being rolled out", though, and failing to meet the latter threshold before rolling them out commercially is imo negligence and even at best simply isn't making the roads safer. To my knowledge they haven't been rolled out anywhere where there's heavy snow in the winter, and it would be ludicrously evil to roll them out somewhere like that before you had a certain ability to handle that common weather condition with a certain degree of safety.
Yes, training them to handle these circumstances is a difficult task, but that difficulty doesn't absolve the companies from ensuring that their vehicles are operating safely where they've been rolled out. This isn't just a matter of "slightly buggier cars" -- being unable to handle common severe weather conditions in an area endangers customers and bystanders. If your goal is to actually increase safety by replacing fallible human drivers, you can't ignore huge safety defects in self-driving cars. The fact that this is the case is why they're backing out of these locations now until they can find solutions to flooded roads that are safe enough, and they should have had better testing in extreme conditions like this to recognize this problem earlier, before rolling out the paid taxi service to consumers.
This is a "pause" while severe weather is likely. The comparison is between driving in such conditions or not, rather than whether they will be in Atlanta or not. It seems similar to how airlines cancel flights when the weather looks too bad?
They are relying on weather forecasts to avoid driving in severe weather, but it turns out they needed to be more conservative about that, as well as improving their cars' ability to cope when the weather suddenly turns bad.
I'm not sure I follow -- Do you mean by reducing the rate of accidents, through reducing the risk of human driving error? Or some other metric?
Yes, as I see it every trip taken in a Waymo reduces the odds of an injury. I’ve also seen them force nearby human drivers to be safer. They drive slowly and can see around corners better than humans. A Waymo showing down near an alley or intersection lets drivers know something is approaching from out of sight.
After the grand-parent comment on blocking ambulance bay, I almost mis-read your comment to mean we have the same number of injuries, but fewer of them survive long enough to arrive at the ER because of self driving cars.
But you meant to say that self driving cars are much safer than human driven cars.
Ignorable aside to explain personal bias: I do wonder if this is because we're having fewer accidents per million cars on the road? Or that accidents are less fatal over same period or over same number accidents? Or that we have fewer cars and hence fewer accidents? Can these factors be attributed to anything else such as high gas prices or people working from home or drinking/partying/driving drunk less? The onus is on me to look up this information and convince myself again and again every time this comes up, so please feel free to ignore.
I will admit my bias is anti big tech and I'm very skeptical of how quickly these things have been approved to be on the road with humans. Eg, we still don't even have women crash test dummies after decades but this we approved this fast? I'm sure that big tech will release all kinds of
lies, damn lies andstatistics skewed to show them in a good light.I can only imagine how robotaxis will behave once deployed to a place with chronic snow 30% of the year. Seems like our roads have no end of edge cases for machine vision with no underlying understanding of the world. Of course human drivers are imperfect, but it's interesting to consider that most humans won't make this mistake as long as they realize that water depth is very difficult to judge. Whereas the robotaxis seem to just... not perceive the water? I wonder why LIDAR doesn't tip them off.
Bodies of water absorb LiDAR, so no light back to the sensor and it will just report a void there. Sometimes you'll get the odd return, particularly on the ground under the water for shallow water. I see it pretty often in my daily work. You'll need something else to sense it.
How often during regular trips do these robotaxis get a lidar void and still drive ahead? I’m suddenly imagining a cartoon style driving off a cliff, pause in midair, and then plummeting with that “oh no not again” look on the cartoon face
Stopping erroneously is, while much safer than driving into a flood, not ideal. And the cars can't stop unconditionally for puddles, or any schmuck with a garden hose can DoS them. (Furthermore: and make it look like an accident.)
To be fair here, "don't drive into floods" is unironically challenging for human drivers, as well. It's tough to judge the depth of a body of water, so a great deal of the "is this a dangerous flood or a shallow puddle" decision comes down to context and heuristics. What's your extrapolated topography of the street? If the water is flowing, what does it look like? Have any other cars gone through it? Is there a schmuck with a garden hose guffawing next to the road? Etc.
There is definitely a reason these trials have heretofore been in socal or the southwest, though.
Waymo maps the roads where they will be driving. I imagine they could predict where flooding might happen based on things like elevation to bridge underpasses, erc. Also, the cars could compare current conditions to what the road looks like when it's dry.
It will be a project, but it seems doable?
Pretty huge project that relies on up-to-date highly-detailed topography. I'll be interested to see how they solve this problem!
Ah our work is almost the total opposite of a robotaxi, we collect it from a flight and process that. Not real-time or solely machine analyzed. Void would be wherever there is open air, so going towards flooding, it would look like a cliff in lidar. Even under less severe flooding, it sounds like they're still panicking after "driving off the cliff". I don't think it'd happen often during normal trips, probably why Waymo didn't already have a solution prepared (other than the total lack of regard for safety).
If you would describe water to an alien who has never seen it, would they be able to identify it? It’s extremely difficult to identify by computers. Some types of sensors might perceive floodwater as solid ground.
It’s this kind of thing which is the reason why I was against the concept of building out self driving cars that work like this instead of on closed purpose-made roads. Computers are really bad at identifying hazards and making complex decisions. It’s practically a miracle that Waymo has a self driving platform that operates as good as it does right now.
I think that discredits the work of the engineers. At some point you have to admit the injury rate is low because they've done an amazing job and has nothing to do with luck. It seems that, for now, they should stick to places with famously easy weather (SF, LA, Phoenix). But they'll figure it out eventually. I've been on team self-driving for decades now. Humans have only been driving because we haven't had a better alternative. I look forward to a future where someone driving a car is viewed as incredibly selfish.
Something can be a miracle of engineering as well!
I look forward to a day when we don’t have absurd amounts of private cars with human drivers in them.
This is one of the touted uses of Google's "world model", Genie. They're using it to teach their cars about unlikely conditions: large animals on the road, conflagrations, tornadoes, and indeed flooding.
They show some of the simulations here, but it's a heavy page. Best viewed on desktop.
Maybe LIDAR has weird reflections off water? Whatever it is, I imagine they will take a few weeks or months to fix this and it will stay fixed. It doesn’t seem like it would be as difficult as kangaroos.
The one benefit I can see to this is it might place the correct stigma on driving during bad weather. While humans can probably navigate heavy snow or rain compared to a "robotaxi", should we?
Sometimes you have no choice -- for instance, if I wouldn't drive in snow, I would literally never leave my house some weeks! And plenty of in-person jobs basically require employees to take on that risk (some fairly, like firefighters, some less so, like in-person office gigs). But I do generally agree that we underestimate the risk!
Also, there is something to be said about prepardness of the region.
An inch of snow is catastrophic to Texas, but in the northeast there might be a 2 hour delay if the schools are feeling generous.
Waymo expands pause to four cities as robotaxis keep driving into floods
From the article:
[...]