Self-driving cars have got a kangaroo problem
August 27, 2024 12:56 AM Subscribe
Self-driving cars have got a kangaroo problem. Self-driving cars are a game changer for disabled Australians, but they've got a kangaroo problem. Darren waits hours and hours and hours for taxis in Queensland, and hopes self-driving cars might be the answer. But an unpredictable marsupial is standing in the way.
I wonder how different kangaroo and deer are, in terms of road danger. (There are a lot of hits (hehe) for searches on Tesla-deer collisions, but they don't really mention whether the cars were in autopilot or not.)
posted by mittens at 1:41 AM on August 27
posted by mittens at 1:41 AM on August 27
Ugh. Musk's rhetoric around Tesla and FSD should be grounds for criminal charges. They're not a serious contender in this space, and they have serious amounts of blood on their hands.
Competent self driving companies are able to deal with problems like this; this is a "Tesla FSD bad" not "self driving hard" problem.
posted by constraint at 1:59 AM on August 27 [7 favorites]
Competent self driving companies are able to deal with problems like this; this is a "Tesla FSD bad" not "self driving hard" problem.
posted by constraint at 1:59 AM on August 27 [7 favorites]
In all fairness, human driven cars also have a kangaroo problem. They're plentiful, a mass of solid muscle, they're nocturnal and emerge as dusk as the light wanes, and any encounter between one and a car usually results in a writeoff.
posted by His thoughts were red thoughts at 2:01 AM on August 27 [7 favorites]
posted by His thoughts were red thoughts at 2:01 AM on August 27 [7 favorites]
It's nice to see some more positive / optimistic reporting about driverless cars! That said, it seems to me that "predictability" shouldn't be the metric for avoiding living creatures (or debris, or large rocks or stopped vehicles) in the roadway. You scan something in the way or on the verge and you slow to a crawl or even stop (giving the car behind you an opportunity to test their own conception of "predictability").
The thing about kangaroos is that
a) because they're bounding along on two spring loaded legs plus a muscular tail, they can change direction very abruptly/very suddenly in a way that deer/horses don't
b) kangaroos are moving very, very fast. Kangaroos can move at speeds of up to 70 km/h (43 mph) over short distances, while they can sustain a speed of 40 km/h (25 mph) for nearly 2 km (1.2 mi).
posted by chariot pulled by cassowaries at 2:08 AM on August 27 [19 favorites]
The thing about kangaroos is that
a) because they're bounding along on two spring loaded legs plus a muscular tail, they can change direction very abruptly/very suddenly in a way that deer/horses don't
b) kangaroos are moving very, very fast. Kangaroos can move at speeds of up to 70 km/h (43 mph) over short distances, while they can sustain a speed of 40 km/h (25 mph) for nearly 2 km (1.2 mi).
posted by chariot pulled by cassowaries at 2:08 AM on August 27 [19 favorites]
Specifically re kangaroos: what do human drivers do to avoid them - and how successful is that? I have heard about car hire companies in Australia limiting or banning night time driving specifically for this reason. Is that correct?
If people are specifically interested in "How safe is safe?" with respect to autonomous driving reliability levels -I would recommend the talk linked here.
posted by rongorongo at 3:45 AM on August 27
If people are specifically interested in "How safe is safe?" with respect to autonomous driving reliability levels -I would recommend the talk linked here.
posted by rongorongo at 3:45 AM on August 27
🦘
posted by HearHere at 3:49 AM on August 27 [1 favorite]
posted by HearHere at 3:49 AM on August 27 [1 favorite]
There's also the whole "we haven't driven hundreds of thousands of hours of driving on the other side of the road" - I'm sure from Elon's point if view it's simply "we invert the x co-ords" but it's not really that simple in the real world
posted by mbo at 4:45 AM on August 27 [1 favorite]
posted by mbo at 4:45 AM on August 27 [1 favorite]
'Unpredictable' kangaroos a roadblock'
In other words, "our autopilot works and is perfectly safe as long as nothing unexpected is encountered." Don't worry, the technology will be ready in five years.
"Often, taxi drivers will see that you're in a wheelchair and scoot around because they will get more money from a whole packed taxi.
"If there was a robotaxi that would just rock up and I drive my wheelchair into it, that would be completely game changing."
Uh, dude, robotaxis would also prefer to make more money. Technology hasn't solved capitalism, my brother.
posted by AlSweigart at 4:48 AM on August 27 [11 favorites]
In other words, "our autopilot works and is perfectly safe as long as nothing unexpected is encountered." Don't worry, the technology will be ready in five years.
"Often, taxi drivers will see that you're in a wheelchair and scoot around because they will get more money from a whole packed taxi.
"If there was a robotaxi that would just rock up and I drive my wheelchair into it, that would be completely game changing."
Uh, dude, robotaxis would also prefer to make more money. Technology hasn't solved capitalism, my brother.
posted by AlSweigart at 4:48 AM on August 27 [11 favorites]
"We are working on creating a remote operation centre in Brisbane, and once that is ready, we will be trying out driving our prototype [car] without a driver in the driver's seat," he says.
It limits the driver's vision, removes their direct incentive for safety, AND it's a cost saving measure.
I foresee problems.
"Five or 10 years ago I'd be on a panel with a bureaucrat or politician and I'd be cringing at every second thing that they said about autonomous vehicles," Professor Milford jokes.
"Nowadays, it's much more common that I can just sit back and listen to what they say, because they're making a lot more sense – I think there's been a really big improvement across the board in that area."
We've been making the "the tech will be ready in five years" jokes for so long, they now try to get in front of it in their advertisement-masquerading-as-journalism pieces now.
posted by AlSweigart at 4:58 AM on August 27 [2 favorites]
It limits the driver's vision, removes their direct incentive for safety, AND it's a cost saving measure.
I foresee problems.
"Five or 10 years ago I'd be on a panel with a bureaucrat or politician and I'd be cringing at every second thing that they said about autonomous vehicles," Professor Milford jokes.
"Nowadays, it's much more common that I can just sit back and listen to what they say, because they're making a lot more sense – I think there's been a really big improvement across the board in that area."
We've been making the "the tech will be ready in five years" jokes for so long, they now try to get in front of it in their advertisement-masquerading-as-journalism pieces now.
posted by AlSweigart at 4:58 AM on August 27 [2 favorites]
> "move fast and break things" = bad idea irl
posted by HearHere at 6:55 AM on August 27 [4 favorites]
posted by HearHere at 6:55 AM on August 27 [4 favorites]
they can change direction very abruptly/very suddenly in a way that deer/horses don't
That is not my experience with deer at all. They can just as quickly change direction on a dime.
posted by jmauro at 7:23 AM on August 27 [2 favorites]
That is not my experience with deer at all. They can just as quickly change direction on a dime.
posted by jmauro at 7:23 AM on August 27 [2 favorites]
> a) because they're bounding along on two spring loaded legs plus a muscular tail, they can change direction very abruptly/very suddenly in a way that deer/horses don't
Deer are amazing at changing direction - they aren't horses. They prance. They can leap 10 meters horizontally. They move at 48-60 kph.
As a driver in deer country, you don't try to swerve to avoid a deer. You slow down. They can and will change direction faster than any swerving you can manage and in a completely unpredictable direction, as they are treating the car as a predator who is trying to eat them.
Based off what I can find, humans killed by deer-collisions and by kangaroo-collisions in Canada vs Australia is pretty similar in order of magnitude. (~20 of people per country per year)
Mass-wise, Deer and Kangaroo are pretty similar.
Deer aren't like horses; they don't move like horses, they don't mass like horses. Moose are more like horses.
posted by NotAYakk at 7:25 AM on August 27 [5 favorites]
Deer are amazing at changing direction - they aren't horses. They prance. They can leap 10 meters horizontally. They move at 48-60 kph.
As a driver in deer country, you don't try to swerve to avoid a deer. You slow down. They can and will change direction faster than any swerving you can manage and in a completely unpredictable direction, as they are treating the car as a predator who is trying to eat them.
Based off what I can find, humans killed by deer-collisions and by kangaroo-collisions in Canada vs Australia is pretty similar in order of magnitude. (~20 of people per country per year)
Mass-wise, Deer and Kangaroo are pretty similar.
Deer aren't like horses; they don't move like horses, they don't mass like horses. Moose are more like horses.
posted by NotAYakk at 7:25 AM on August 27 [5 favorites]
If kangaroos and deer are a problem, moose more so. Moose are a half-tonne of solid muscle with a high centre of gravity. They're the reason that New Brunswick highways have massive fences, and also why the province has one of the highest demands for paraplegic care.
posted by scruss at 7:40 AM on August 27 [3 favorites]
posted by scruss at 7:40 AM on August 27 [3 favorites]
Hitting a moose will very possibly send the entire thing through your front window into the body of your car, a situation the Swedes call "the meat grinder".
posted by Iteki at 7:46 AM on August 27 [1 favorite]
posted by Iteki at 7:46 AM on August 27 [1 favorite]
The article is a little odd; I kind of wished they had separated the first part (usefulness of true autopilot to disabled people) from the rest (problems with self-driving) and maybe a third (specific problems with kangaroos). It was interesting, but I felt a little whiplash as the article move from one topic to another.
posted by GenjiandProust at 8:14 AM on August 27 [2 favorites]
posted by GenjiandProust at 8:14 AM on August 27 [2 favorites]
robotaxis would also prefer to make more money. Technology hasn't solved capitalism, my brother.
It heartbreaking that people can't see that even if robot taxis could accomplish everything they promise; wide adoption by companies that got their start flouting regulation would result in a system that would rapidly enshittify to something worse than what we have now with a veneer of it's not our fault it's the algorithm.
Plus no resiliency.
posted by Mitheral at 8:43 AM on August 27 [5 favorites]
It heartbreaking that people can't see that even if robot taxis could accomplish everything they promise; wide adoption by companies that got their start flouting regulation would result in a system that would rapidly enshittify to something worse than what we have now with a veneer of it's not our fault it's the algorithm.
Plus no resiliency.
posted by Mitheral at 8:43 AM on August 27 [5 favorites]
In all fairness, human driven cars also have a kangaroo problem.
In all fairness, the problem isn't kangaroos. Kangaroos are just the latest incarnation of the same old problem: how to program the car to act reasonably in the bajillions of variations of previously unencountered situations. This is a problem humans can generally be trusted to handle and algorithms cannot. And maybe cannot ever be.
Tesla and other companies aren't interested in solving that problem as much as the problem of escaping culpability for their product failures:
A NHTSA report on its investigation into crashes in which Tesla vehicles equipped with the automaker's Autopilot driver assistance feature hit stationary emergency vehicles has unearthed a troubling detail: In 16 of those crashes, "on average," Autopilot was running but "aborted vehicle control less than one second prior to the first impact."
posted by AlSweigart at 9:07 AM on August 27 [7 favorites]
In all fairness, the problem isn't kangaroos. Kangaroos are just the latest incarnation of the same old problem: how to program the car to act reasonably in the bajillions of variations of previously unencountered situations. This is a problem humans can generally be trusted to handle and algorithms cannot. And maybe cannot ever be.
Tesla and other companies aren't interested in solving that problem as much as the problem of escaping culpability for their product failures:
A NHTSA report on its investigation into crashes in which Tesla vehicles equipped with the automaker's Autopilot driver assistance feature hit stationary emergency vehicles has unearthed a troubling detail: In 16 of those crashes, "on average," Autopilot was running but "aborted vehicle control less than one second prior to the first impact."
posted by AlSweigart at 9:07 AM on August 27 [7 favorites]
I've never really had a problem with deer changing directions. Deer are unpredictable because they're hiding in the trees and bushes, where you can't even see them until they jump in front of you. Slowing down around the visible deer is good enough; it's the invisible ones that get ya.
posted by I-Write-Essays at 9:14 AM on August 27 [1 favorite]
posted by I-Write-Essays at 9:14 AM on August 27 [1 favorite]
Deer problems are solved with non-autonomous drivers by attaching a grill guard on the front of the vehicle to ram them out of the way without any damage to the vehicle and without having to slow down where they are present, so sorry about the deer lost.
So if self-driving cars can solve the deer or kangaroo problem via safer methods for all involved that would be a great thing.
posted by The_Vegetables at 10:31 AM on August 27
So if self-driving cars can solve the deer or kangaroo problem via safer methods for all involved that would be a great thing.
posted by The_Vegetables at 10:31 AM on August 27
Tesla FSD isn't perfect, but it works damn good. I'm a pretty safe driver - no accidents in the last 25 years - and it definitely is a safer driver than me.
A lot of the agita about self-driving cars is fueled not by good-faith skepticism of the technology but out of recognition that robust self-driving technology will end demand for public transportation other than by the poor.
posted by MattD at 1:03 PM on August 27
A lot of the agita about self-driving cars is fueled not by good-faith skepticism of the technology but out of recognition that robust self-driving technology will end demand for public transportation other than by the poor.
posted by MattD at 1:03 PM on August 27
Well, except that self-driving cars will do less than nothing to reduce congestion and, if anything, require more road infrastructure to support. We need fewer cars, not better cars.
posted by GenjiandProust at 1:10 PM on August 27 [4 favorites]
posted by GenjiandProust at 1:10 PM on August 27 [4 favorites]
Maybe we really need to just give kangaroos cars; that should sort everything out.
posted by GenjiandProust at 1:17 PM on August 27 [7 favorites]
posted by GenjiandProust at 1:17 PM on August 27 [7 favorites]
"aborted vehicle control less than one second prior to the first impact."
(ah yes, the screaming Red Steering Wheel of Impending Doom)
Technically that's still ADAS "Level 3" as defined by the SAE . . .
Level 4 is you don't need driver controls but the car can't go everywhere (or necessarily much of anywhere), and Level 5 is full human-removing ADAS.
SAE missed a step, there should be a Level 3.5 where the car hands over to the human driver with sufficient time to not crash etc.
Level 3 Autonomy Is Confusing Garbage
Tesla's A/P has indeed killed more than a few inattentive drivers, as it's pretty easy to trust it with your life since it seems to work really great, until it doesn't.
"FSD" is another kettle of fish that is very hard for me to analyze and/or critique. Basically the whole program is not the approach I would have proceeded had I been given the ball 10 years ago, e.g. job one would have been basic lane-keeping and not hitting or driving over things in the car's path, but that's not been much of a priority for Tesla for some reason.
posted by torokunai at 1:34 PM on August 27 [2 favorites]
(ah yes, the screaming Red Steering Wheel of Impending Doom)
Technically that's still ADAS "Level 3" as defined by the SAE . . .
Level 4 is you don't need driver controls but the car can't go everywhere (or necessarily much of anywhere), and Level 5 is full human-removing ADAS.
SAE missed a step, there should be a Level 3.5 where the car hands over to the human driver with sufficient time to not crash etc.
Level 3 Autonomy Is Confusing Garbage
Tesla's A/P has indeed killed more than a few inattentive drivers, as it's pretty easy to trust it with your life since it seems to work really great, until it doesn't.
"FSD" is another kettle of fish that is very hard for me to analyze and/or critique. Basically the whole program is not the approach I would have proceeded had I been given the ball 10 years ago, e.g. job one would have been basic lane-keeping and not hitting or driving over things in the car's path, but that's not been much of a priority for Tesla for some reason.
posted by torokunai at 1:34 PM on August 27 [2 favorites]
Specifically re kangaroos: what do human drivers do to avoid them - and how successful is that?
Slowing down helps, otherwise a big bull bar seems to be a popular solution. There’s a reason why they say that the first kangaroo most tourists see is dead by the side of the road.
posted by thixotemperate at 1:38 PM on August 27 [4 favorites]
Slowing down helps, otherwise a big bull bar seems to be a popular solution. There’s a reason why they say that the first kangaroo most tourists see is dead by the side of the road.
posted by thixotemperate at 1:38 PM on August 27 [4 favorites]
Clearly the solution to this problem is to tie me kangaroos down, sport, tie me kangaroos down
posted by InfidelZombie at 2:07 PM on August 27 [6 favorites]
posted by InfidelZombie at 2:07 PM on August 27 [6 favorites]
A lot of the agita about self-driving cars is fueled not by good-faith skepticism of the technology but out of recognition that robust self-driving technology will end demand for public transportation other than by the poor.
If you shut down public transit in major cities tomorrow and gave everyone a self-driving car, no one would be able to get to work because individual cars take up way more space than transit to move the same number of people. So all the self-driving cars would be parked in the middle of an enormous traffic jam, auto-honking at each other when they get too close.
posted by hydropsyche at 2:17 PM on August 27 [5 favorites]
If you shut down public transit in major cities tomorrow and gave everyone a self-driving car, no one would be able to get to work because individual cars take up way more space than transit to move the same number of people. So all the self-driving cars would be parked in the middle of an enormous traffic jam, auto-honking at each other when they get too close.
posted by hydropsyche at 2:17 PM on August 27 [5 favorites]
If you think about it, it's insane we still rely on steering wheels and pedals to control vehicles. I currently drive my van with a joystick and a touch screen due to my disability.
I was hopeful my current vehicle (2015) would be the last traditional vehicle I'd own but it doesn't feel like that anymore to me. Self driving was overhyped by people who forgot about the fine details. I am happy the tech has moved forward, though. I also live in an area where there are zero accessible taxis (a 20 minute ride to the airport in a non-emergency accessible transit vehicle is $150, not even joking). So maybe, just maybe when I can no longer drive with the aide of significant adaptive equipment ($75,000) there will be a fully self-driving fleet awaiting my order but I'm not going to hold my breath.
posted by thorny at 3:15 PM on August 27 [6 favorites]
I was hopeful my current vehicle (2015) would be the last traditional vehicle I'd own but it doesn't feel like that anymore to me. Self driving was overhyped by people who forgot about the fine details. I am happy the tech has moved forward, though. I also live in an area where there are zero accessible taxis (a 20 minute ride to the airport in a non-emergency accessible transit vehicle is $150, not even joking). So maybe, just maybe when I can no longer drive with the aide of significant adaptive equipment ($75,000) there will be a fully self-driving fleet awaiting my order but I'm not going to hold my breath.
posted by thorny at 3:15 PM on August 27 [6 favorites]
I wonder how different kangaroo and deer are, in terms of road danger.
You're not going to get into a fistfight with a super yoked deer.
posted by snuffleupagus at 5:12 PM on August 27 [2 favorites]
You're not going to get into a fistfight with a super yoked deer.
posted by snuffleupagus at 5:12 PM on August 27 [2 favorites]
Transit advocates actually understand how much most Americans hate transit and love riding in their own cars. They know that suburban to core commercial district commutes are regularly over an hour door to door despite being well under 30 road miles. This means that you beat transit by going 25 miles per hour on average or less. When you consider the hedonic benefit of going from the front door of your house to the door of your office building in air conditioned solitude reading or watching a movie, you'll permit 15 or 20 minutes more - maybe 20 miles per hour or less. Lots of tolerance of traffic jams while waiting for the computers to figure those out too.
posted by MattD at 10:31 PM on August 27 [2 favorites]
posted by MattD at 10:31 PM on August 27 [2 favorites]
Transit advocates actually understand how much most Americans hate transit and love riding in their own cars.
They know that because the spending on transit, even in the Biden budget (a guy who supports transit) is less than 5% of the highway expansion budget. It's like knowing that people prefer eating lobster and steak to McDonalds because the government subsidizes lobsters such that they cost .20 cents.
posted by The_Vegetables at 8:12 AM on August 28 [3 favorites]
They know that because the spending on transit, even in the Biden budget (a guy who supports transit) is less than 5% of the highway expansion budget. It's like knowing that people prefer eating lobster and steak to McDonalds because the government subsidizes lobsters such that they cost .20 cents.
posted by The_Vegetables at 8:12 AM on August 28 [3 favorites]
I say this as someone who didn't drive for years and would enjoy some non-drive knitting time: literally you can get almost anywhere faster if you do it yourself in a car, even in heavy traffic. Taking busses, trains, etc. can be an hour to three hours or worse at times, and they have limited options for where to go and where to stop and where to pick up. That's why Americans "hate public transit."
posted by jenfullmoon at 9:36 AM on August 28 [4 favorites]
posted by jenfullmoon at 9:36 AM on August 28 [4 favorites]
In my experience the problem with roos is not speed but acceleration. They'll be chilling out at the side of the road and at the last second decide to launch themselves in front of a vehicle, or into the side of it.
Thermal imaging might be the go to help the robots "see" them hiding in the bush.
posted by onya at 1:12 PM on August 28 [1 favorite]
Thermal imaging might be the go to help the robots "see" them hiding in the bush.
posted by onya at 1:12 PM on August 28 [1 favorite]
In my experience the problem with roos is not speed but acceleration. They'll be chilling out at the side of the road and at the last second decide to launch themselves in front of a vehicle, or into the side of it.
I found quite a lot of dashcam footage, which I’ll choose not to share here, which supports this. Driving at the sort of speed one might consider sensible on a straight, wide country road - there might be only about 1 second between catching sight of a roo and hitting it.
posted by rongorongo at 2:51 PM on August 28 [1 favorite]
I found quite a lot of dashcam footage, which I’ll choose not to share here, which supports this. Driving at the sort of speed one might consider sensible on a straight, wide country road - there might be only about 1 second between catching sight of a roo and hitting it.
posted by rongorongo at 2:51 PM on August 28 [1 favorite]
> If you shut down public transit in major cities tomorrow and gave everyone a self-driving car, no one would be able to get to work because individual cars take up way more space than transit to move the same number of people
Ok, NACTO:
https://nacto.org/publication/transit-street-design-guide/introduction/why/designing-move-people/
I hate people supporting arguments whose conclusion I want to believe with horrible graphs. "People per hour" is not a useful metric - "people per km per hour" is. We care about throughput, not capacity.
You can have 9000 people walking over a wide piece of sidewalk, but they are moving only 4 km in that hour. So the bandwidth is 36,000 people-km. The same width as asphalt has the cars going 60 km in an hour, so the 1000 cars at 60 km per hour is 60 kpkm (1000 people kms).
Bikers in dense urban areas maybe move 15 km/hour, so you get 112 kpkm.
A bus might average 40 km/hour (due to having to stop) in a densely used transit lane; at 5000 people that is 200 kpkm. A rail system hits approx 600 kpkm.
The infographic in that link, meanwhile, shows sidewalks above dedicated transit lines.
And the fact they think this infographic is at all useful - that it passes the sniff test at all - makes me not believe any conclusions they make about this subject. Either their judgement is garbage or they are making decisions that are dishonest. Is their bikeway throughput accurate? Determining that is work, but based on their decision about an infographic comparing sidewalk to transit lane capacity, I should assume it isn't accurate or reflective of anything useful.
posted by NotAYakk at 8:11 AM on August 29
Ok, NACTO:
https://nacto.org/publication/transit-street-design-guide/introduction/why/designing-move-people/
I hate people supporting arguments whose conclusion I want to believe with horrible graphs. "People per hour" is not a useful metric - "people per km per hour" is. We care about throughput, not capacity.
You can have 9000 people walking over a wide piece of sidewalk, but they are moving only 4 km in that hour. So the bandwidth is 36,000 people-km. The same width as asphalt has the cars going 60 km in an hour, so the 1000 cars at 60 km per hour is 60 kpkm (1000 people kms).
Bikers in dense urban areas maybe move 15 km/hour, so you get 112 kpkm.
A bus might average 40 km/hour (due to having to stop) in a densely used transit lane; at 5000 people that is 200 kpkm. A rail system hits approx 600 kpkm.
The infographic in that link, meanwhile, shows sidewalks above dedicated transit lines.
And the fact they think this infographic is at all useful - that it passes the sniff test at all - makes me not believe any conclusions they make about this subject. Either their judgement is garbage or they are making decisions that are dishonest. Is their bikeway throughput accurate? Determining that is work, but based on their decision about an infographic comparing sidewalk to transit lane capacity, I should assume it isn't accurate or reflective of anything useful.
posted by NotAYakk at 8:11 AM on August 29
I want to move to your city where car traffic averages 60km/h during peak hour.
posted by onya at 7:22 PM on August 29 [1 favorite]
posted by onya at 7:22 PM on August 29 [1 favorite]
Also I'm not sure that your assertion that throughput is the only useful metric? If you have a sporting event end what you really care about is the rate at which you can get people out of there and on their way home and less about how long their total journey will be.
A bullet train might have great "kpkm" but that doesn't make it the ideal way for every single person to commute to work.
posted by onya at 7:45 PM on August 29 [1 favorite]
A bullet train might have great "kpkm" but that doesn't make it the ideal way for every single person to commute to work.
posted by onya at 7:45 PM on August 29 [1 favorite]
« Older "Woof. Nate. Not cool man." | We work to create information that we will never... Newer »
posted by chavenet at 1:28 AM on August 27 [4 favorites]