Volvo's Driverless Cars 'Confused' by Kangaroos (bbc.com)

An anonymous reader shares a report: Volvo's self-driving technology is struggling to identify kangaroos in the road. The Swedish car-maker's 2017 S90 and XC90 models use its Large Animal Detection system to monitor the road for deer, elk and caribou. But the way kangaroos move confuses it. "We've noticed with the kangaroo being in mid-flight when it's in the air, it actually looks like it's further away, then it lands and it looks closer," its Australia technical manager said. But the problem would not delay the rollout of driverless cars in the country, David Pickett added.

Comments

Kangaroo vs White-Tailed Deer

[damn_registrars]

The White-Tailed Deer (and likely other related species - I'm only personally familiar with this one) moves in a bounding manner, often with all four legs off the ground. What makes it that much different from the kangaroo in the eyes of the driverless car? Is it at in the height difference in terms of on the ground versus not on the ground?

Re:Kangaroo vs White-Tailed Deer

[OrangeTide]

The difference is the lack of training data. Volvo didn't try hard enough to get this right.

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

Deer only bound like that occasionally, its not their default locomotion,. Kangaroos hop as their primary means of movement and they can bound easily 20 feet in distance and 5-6 feet in height and do so on a pretty regular basis. In short, what makes the deer look different than a kangaroo to a car? Pretty much everything.

Here is another question, does a deer look anything like a kangaroo to you?

Re:Kangaroo vs White-Tailed Deer

[religionofpeas]

I expect that Volvo did most of the training with animals they find on the road in Sweden. The White Tailed Deer lives in North America.

Re:Kangaroo vs White-Tailed Deer

[David_Hart]

The difference is the lack of training data. Volvo didn't try hard enough to get this right.

I'm guessing that the system assumes that any object in it's path is on the ground. Anything in the air would appear to be further away as the ground would only rise up to meet it in the distance. Basically, they didn't throw bouncy balls at the car to see what would happen....

PS: One of the US ads for car safety includes signs indicating that bouncing balls have kids following them.

Re:Kangaroo vs White-Tailed Deer

[DontBeAMoran]

Volvo isn't in the habit of testing their cars upside-down.

Re:Kangaroo vs White-Tailed Deer

[DontBeAMoran]

Does a deer look anything like a kangaroo to you?

Only when I'm drunk.

Re:Kangaroo vs White-Tailed Deer

[Dutch Gun]

I know it's passé to read TFA, but:

Volvo's safety engineers began filming kangaroos' roadside behaviour in a nationally recognised hotspot for collisions in 2015.

There's a picture of a Volvo vehicle with "Kangaroo detection data collection vehicle" printed on its side.

So... they're working on it, with good training data.

Re: Kangaroo vs White-Tailed Deer

[Anonymous Coward]

We have plenty of varieties of deer in Sweden. Some of which do bound.

The body shape of a kangaroo is quite different from other animals though.

Either way. Seems like this indicates a need to analyse the sensor data from this type of source to identify it properly in the future. Or the need for an extra sensor type.

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

You stupid fucktard. This article IS the traing data results specifically for kangaroos. The picture of the car has LARGE letters that say KANGAROO DETECTION DATA COLLECTION VEHICLE.

Re: Kangaroo vs White-Tailed Deer

[Wycliffe]

Either way. Seems like this indicates a need to analyse the sensor data from this type of source to identify it properly in the future. Or the need for an extra sensor type.

If the shape of the object interferes with your code then you're doing it wrong. Whether it is a kangaroo, a bird, a box, a balloon, or a sheet of glass, if something is in the roadway, you need to stop. Because of the lack of intelligence and depth perception, I would expect a self driving car to pretty much require a second sensing mechanism like sonar, radar, etc...

Re:Kangaroo vs White-Tailed Deer

[dj245]

I expect that Volvo did most of the training with animals they find on the road in Sweden. The White Tailed Deer lives in North America.

But Moose live in Sweden. Moose are much larger than deer, but share the same basic body type (large body on relatively long and thin legs). Special caution needs to be taken with moose, since their eyes don't glow in the dark, and they have tendencies to bite. In fact, a moose once bit my sister.

Re: Kangaroo vs White-Tailed Deer

[Anonymous Coward]

I have worked professionally with the kind of system used in self-driving cars.
The trouble is that usually the video frames are analysed as static images, using no information from the previous frame. Continuity of objects moving around is puzzled together after the fact at a higher software level.
So what really needs to be added here, probably at that higher level because we're just not ready for the alternative beyond ‘toy’ applications and static mapping, is a bit of code that checks the scale estimate. Since objects don't magically get bigger, you can deduce that the object is no longer standing on the ground further away, but has moved up instead.
What would also help is the addition of an extra camera for parallax, or a depth sensor. Various kinds exist, for example using a kind of radar, or a pattern projector.

Re:Kangaroo vs White-Tailed Deer

[wonkey_monkey]

Didn't you know? If driverless cars can't work 100% perfectly now, how will they ever do so in the future?

I mean, that is proper actual logic, that.

Re:Kangaroo vs White-Tailed Deer

[TheCastro1689]

https://i.ytimg.com/vi/ZfxJXBM... Yes, they look very similar.

Re:Kangaroo vs White-Tailed Deer

[rogoshen1]

what about when the deer are drunk, what then?

Re:Kangaroo vs White-Tailed Deer

[DontBeAMoran]

When the deer are drunk, cars look like metal bears.

Re: Kangaroo vs White-Tailed Deer

[Anonymous Coward]

no. just ban kangaroos.

Re:Kangaroo vs White-Tailed Deer

[nnet]

newfoundland speedbump...

Re:Kangaroo vs White-Tailed Deer

[OrangeTide]

I'm guessing that the system assumes that any object in it's path is on the ground.

That's not how these systems work. It's a combination of image processing, Computer Vision and neutral networks. Most of the good ones will stop if a soccer ball bounces through the street too. And if you dig into how the system categorizes pedestrians you'll find that you can trick a poorly trained one if a person doesn't form a neat triangle. Like if they are in a wheelchair or have their leg up in a cast and using crutches. (or a child on a pogo stick)

The human brain organizes these things differently than a Computer Vision system. And you'll have a hard time understanding how it works by comparing the computer to yourself.

Re:Kangaroo vs White-Tailed Deer

[OrangeTide]

Volvo's safety engineers began filming kangaroos' roadside behaviour in a nationally recognised hotspot for collisions in 2015.

No, it's not good training data.

Re:Kangaroo vs White-Tailed Deer

[viperidaenz]

Unless you see kangaroos in a zoo setting. Then they mope around walking like a tripod on their feet and tails without enough room to jump.

Re:Kangaroo vs White-Tailed Deer

[OrangeTide]

Didn't you know? If driverless cars can't work 100% perfectly now, how will they ever do so in the future?

I mean, that is proper actual logic, that.

It's the same people that demand zero insect parts in their wheat and corn. It's just not going to happen. We can keep the amount of crushed up insects you accidentally ingest very low, but it's unreasonable to expect it to be zero. The amount of dead mice (by mass) ground up with your flour is even lower than the insect parts.

PS - The amount of urine in a public swimming pool is much higher than it is reasonable, and everyone has a right to complain about that and demand that it be lower.

Re:Kangaroo vs White-Tailed Deer

[viperidaenz]

I don't think you need to worry too much about a moose biting your car when you're hurtling towards it at speed.

Re:Kangaroo vs White-Tailed Deer

[OrangeTide]

You stupid fucktard. This article IS the traing data results specifically for kangaroos. The picture of the car has LARGE letters that say KANGAROO DETECTION DATA COLLECTION VEHICLE.

I've read TFA and I am familiar with the details beyond TFA. The data is not very good. "The difference is the lack of training data. Volvo didn't try hard enough to get this right."

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

Well, if it takes you a few years to figure out how to recognize one specific kind of obstruction, that doesn't exactly bode well for the thousands of others that are in play. The problem gets exponentially more difficult as you add more types of obstacles and different responses. If it's just a matter of saying "Hey meatbag, look at this," then it's no big deal. Working properly without the meatbag is another matter entirely.

Re:Kangaroo vs White-Tailed Deer

[serviscope_minor]

a moose once bit my sister.

Realli?

Re:Kangaroo vs White-Tailed Deer

[thegarbz]

A lot of the answer is in the vertical movement and the erratic movement. It's not just the legs leaving the ground, a kangaroo when it jumps ... jumps. When sprinting (away from cars especially) they will easily cover their own height in each leap and cover a lot of horizontal ground in the process. When contacting the ground it can change direction suddenly. It's an incredible problem for computer vision to analyse.

Re:Kangaroo vs White-Tailed Deer

[Mr D from 63]

The difference is the lack of training data. Volvo didn't try hard enough to get this right.

Maybe they should just train the damn kangaroos to stay the heck out of the way.

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

Well, if it takes you a few years to figure out how to recognize one specific kind of obstruction, that doesn't exactly bode well for the thousands of others that are in play.

Well, still a better situation than we have for human drivers where every single one has to be manually trained to deal with every situation.
Unfortunately human drivers make misjudgements all the time and drives into stuff instead.
But hey, at least you were killed by a human instead of a machine.

Re: Kangaroo vs White-Tailed Deer

[Anonymous Coward]

I would expect a self driving car to pretty much require a second sensing mechanism like sonar, radar, etc...

Yes, "second" is the keyword here.
Just because they have a second sensing mechanism doesn't mean they can shrug it off.
The first thing about safety critical system is redundancy.
Sometimes systems break or fail, so you need two different systems that operate completely differently, and you need them both to work under normal conditions so that it is unlikely that both fails at the same time.
Their vision system is "confused" by kangaroos. That doesn't mean that the car didn't stop for them, it just means that when the second sensing mechanism breaks the car won't handle kangaroos safely.
In conclusion their tech isn't ready for use in Australia yet but they are working on it.
Nothing really surprising, or exiting for that matter.

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

Whöosh!

Re: Kangaroo vs White-Tailed Deer

[backslashdot]

As more self driving vehicles are on the road the more anomalous situations they will encounter to near miss and learn from.

Self driving cars are still very very rare. Self driving technology is safer than human drivers. We need self driving cars. I hope in the future whoever wants to be a human driver must pass a rigorous exam proving they are at least as safe as an autonomous machine driver. My guess is only a small percent of today's drivers would be able to pass.

Re:Kangaroo vs White-Tailed Deer

[aliquis]

Swedish deer:
http://www.jaktjournalen.se/wp...
And reindeer:
http://jaktmarkerna.se/wp-cont...

Can't be very different?

Re:Kangaroo vs White-Tailed Deer

[ChunderDownunder]

(Sorry to any vegetarians)

Much better to market the vehicles as hunting machines.

With climate change, it'd fix the environment in switching to gamey roo meat rather than continue to destroy the landscape with cattle farming. Cows belch emissions and require land clearing for pasture.

Re:Kangaroo vs White-Tailed Deer

[OutOnARock]

...but he got the Berkowitz's by mistake.....

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

The Australian air force modified their training simulators to model the behavior of groups of wild kangaroos in order to make helicopter training more environmental friendly. They took the AI code of ground troops and replaced the geometry with kangaroos. Following natural behavior, the simulated kangaroos would first run away from the threat object, and seek cover behind bushes or hills. But the AI programmers forgot to take the bits involving regrouping and launching a counter-offensive with anti-aircraft weaponry.

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

They said that about Chess AI, Go AI, numberplate recognition, voice recognition, music recognition, but all of those are now consumer technology.

Avoiding kangaroos, elephants, rhinos, moose, crocodiles, wolves are all region based hazards. Possibly they'll need different AI for countries with right-hand drive, multi-lane roundabouts like the Arc-De-Triomphe in France:

https://vignette3.wikia.nocookie.net/pixar/images/6/6b/Cars_2_arc_de_triomphe_paris.png

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

It will fly through the windscreen and crush the occupants, causing broken hands and arms:

http://www.mirror.co.uk/news/weird-news/carnage-after-moose-hits-car-3846329

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

Old school - In Oz they invented Roo Bars; Kangaroo detection not needed

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

I love pissing in the pool and will never stop doing it.

Re:Kangaroo vs White-Tailed Deer

[MichaelSmith]

And what happens if it sees a bike rider ahead of it but fails to see the wheels?

Re:Kangaroo vs White-Tailed Deer

[goose-incarnated]

I love pissing in the pool and will never stop doing it.

If you stop doing it from the diving board fewer people will complain.

Re:Kangaroo vs White-Tailed Deer

[goose-incarnated]

Didn't you know? If driverless cars can't work 100% perfectly now, how will they ever do so in the future?

I mean, that is proper actual logic, that.

No. The "proper actual logic" (which, in a fit of irony, you appear to miss) is that it's premature to crow about the future existence of SDCs when computational power running SDCs was increased a thousandfold over two decades for a measly 1% increase in performance.

The only people who believe that SDCs are right around the corner are those who are not aware of all the SDC successes of the 90s. The current state of SDCs have them possessing less obstacle avoidance abilities than a cockroach.

For five years people like yourself have been whining that SDCs are only five years away. All the evidence says that the incremental improvements being made have already reached the point of diminishing returns - the extra expenditure of effort has not yet resulted in similar progress.

Re:Kangaroo vs White-Tailed Deer

[Anonymous Coward]

The main problem is that as far as software is concerned Kangaroos are armed with missiles. http://www.snopes.com/humor/nonsense/kangaroo.asp

Re: Kangaroo vs White-Tailed Deer

[Anonymous Coward]

Self driving cars are still very very rare.

Technically true, because they don't exist. They're still in the early stages of development.

Self driving technology is safer than human drivers.

On the micro level, on a case-by-case basis, maybe. On the macro level in aggregate, that has yet to be seen. There are a lot of things that go into safety, you can't just cherry pick a few cases and call one side a definitive winner. Without the total package to evaluate, this is pure speculation. And a lot of the features that contribute to safety can be applied to both cases anyway. They're just expensive.

We need self driving cars.

Accidents in the US are quite rare. Fatalities even more so, only about one every 100 million miles. We perceive things quite differently though because one thing that humans are worse at than driving is processing big numbers. On top of that, we see plenty of "bad" driving that is tolerated without consequence. So human drivers are terrible but that terrible driving almost never causes serious problems, thus failing to provide feedback to correct the human failings while reinforcing a negative impression of human driving that doesn't fit the results.

I hope in the future whoever wants to be a human driver must pass a rigorous exam proving they are at least as safe as an autonomous machine driver. My guess is only a small percent of today's drivers would be able to pass.

Safe in what way? Safely able to operate in a real-world environment without causing damage at a rate above statistical noise? That's basically where most human drivers are at now. I'm all for revoking the licenses of the more flagrant violators, so sure, make them ride in hypothetical self-driving cars. But once you remove the outliers, you're going to be left with a safety standard that even a perfect self-driving setup would be hard-pressed to match. You can teach a car to do all the neat tricks you want, but it still has to work in a reality that involves 3 trillion vehicle miles traveled per year in the US alone.

Re:Kangaroo vs White-Tailed Deer

[david_thornley]

My current non-self-driving car has better obstacle avoidance abilities than a cockroach. Of course, having an allegedly intelligent driver adds additional safety features.

Re:Kangaroo vs White-Tailed Deer

[scdeimos]

I'm guessing that the system assumes that any object in it's path is on the ground.

That's not how these systems work.

I think you'll find it is. They're not using stereoscopic camera systems so they can't directly measure the distance of an object in view.

In steroscopic systems a Difference Map allows you to estimate the distance by, in conjunction with the optic parameters, measuring the number of pixels offset between the left and right image any given object is.

In monoscopic camera systems, such as what Volvo's using, regardless of Neural Network magic, you can only estimate an object's distance by its vertical position in the image based on some calculated (or calibrated) horizon line setting "infinite" and working back from there. The further up from the bottom it is the further away it must be and vice-versa. As David_Hart said, these kinds of systems are brain dead and cannot reliably measure the distance of any objects suspended above the ground.

Re: Kangaroo vs White-Tailed Deer

[swamp_ig]

You will have to take my pool piss from my cold dead.... Yeh :)

Re:Kangaroo vs White-Tailed Deer

[RockDoctor]

There's a picture of a Volvo vehicle with "Kangaroo detection data collection vehicle" printed on its side.

So... they're working on it, with good training data.

Maybe they've been using Apple iKangaroos but they're just holding them wrong?

Me too.

[Anonymous Coward]

I'm confused by Kangaroos too.

Re:Me too.

[K. S. Kyosuke]

But you don't need to now that Volvo has invented artificial confusion!

Re:Me too.

[Anonymous Coward]

I don't need to now? what the fuck does that mean, you dimwitted cock-gobbler? Go punch your mom's mud whistle.

Hugs and kisses,

Juan Epstein

Re:Me too.

[Anonymous Coward]

if you're going to use that old joke, the get it right and use Epstein's mom, you insensitive clod

Re:Me too.

[TechyImmigrant]

I'm confused by Kangaroos too.

Let's not jump to conclusions.

Re:Me too.

[K. S. Kyosuke]

Good job quoting six fifths of a clause. :)

Re:Me too.

[Anonymous Coward]

what old joke?

Re:Me too.

[tsqr]

Were you really in such a hurry to troll that you couldn't apply some basic reading comprehension, or did you comprehend but were just overwhelmingly infuriated at his omission of what would have been a helpful comma after "to"?

Re:Me too.

[Anonymous Coward]

Go stretch yourself, neckbeard.

Re:Me too.

[Anonymous Coward]

Mutant Marsupials Take Up Arms Against Australian Air Force
http://www.snopes.com/humor/nonsense/kangaroo.asp
This seems applicable here.

Re:Me too.

[youngone]

I'm confused by Kangaroos too.

Of course you're confused by Kangaroos. They're 6-foot tall mice that are not even proper mammals, and they keep stuff in a pouch on the front of their fur coat.

I believe they're also keen boxing fans, so keep away from them, especially on Friday nights after the pubs close, they're Australian, so are probably heavy drinkers.

Binocular Vision

[Anonymous Coward]

Use 2 or more cameras to perceive depth? Or have one camera that slowly wobbles left and right.

Re:Binocular Vision

[DontBeAMoran]

In the Animatrix, the machines have more than two eyes.

I don't know why people in A.I./etc are so obsessed with doing everything with a single camera. Those things are cheap these days.

Re:Binocular Vision

[Oswald McWeany]

In the Animatrix, the machines have more than two eyes.

I don't know why people in A.I./etc are so obsessed with doing everything with a single camera. Those things are cheap these days.

Two cameras would probably be smart. Nature evolved to have multiple eyes for a reason (plus it introduces a fail-safe).

With that said. It's probably NOT the cameras that are expensive, but writing code to interpret data from two different cameras and merge it into an interpreted 3D image that is expensive.

Re:Binocular Vision

[religionofpeas]

Even better would be to integrate data from all sensors into one vision: camera + IR + ultrasound + radar + lidar, all around the vehicle, and then construct a unified 3D model of the surroundings.

Re:Binocular Vision

[Greystripe]

Have to wonder how many kangaroo crashes before it would pay for itself.

Re:Binocular Vision

[DontBeAMoran]

Probably still easier than writing code to interpret data from a single camera into an interpreted 3D image.

Re:Binocular Vision

[aaronb1138]

If writing more code to generate true 3D data from two cameras is too much work, then those companies shouldn't be in the automated driving industry.

As I have said many times before, the software must be gnostic, knowing what is ahead and with all sensor suites agreeing on what is ahead within overlapping resolution scopes. This is non-negotiable and should be subject to regulatory scrutiny / testing for level 4-5 automation.

Re:Binocular Vision

[Anonymous Coward]

can't handle one, so more is better?

Re:Binocular Vision

[viperidaenz]

If the kangaroo crashes write the car off and insurance buys the customer a new car, each crash generates profit.

Re:Binocular Vision

[angel'o'sphere]

All self driving/autonomous/driver assisted cars have minimum 2 cameras.
Facepalm.

Re:Binocular Vision

[Anonymous Coward]

If writing more code to generate true 3D data from two cameras is too much work, then those companies shouldn't be in the automated driving industry.

Writing the code is "easy". Proving that it is safe is expensive as hell.

As I have said many times before, the software must be gnostic, knowing what is ahead and with all sensor suites agreeing on what is ahead within overlapping resolution scopes. This is non-negotiable and should be subject to regulatory scrutiny / testing for level 4-5 automation.

OK, that is your opinion.
What actually matters in this case is ISO 26262. Read it and develop the system accordingly.
Yes, it covers every obscure theoretical case that will never happen that has been listed on Slashdot.
Or more specifically, you have to think about them, document that you have thought about them and documented how you avoid that case leading to a dangerous situation.
And yes, you need to document every safety related code path in your software to show that the code can't execute in an unintended order so say goodby to exceptions, garbage collectors, abstractions and everything else you like about your favorite high level language.
You also need to have an idea of how to deal with what happens if a random transistor in the ALU fries and other possible hardware failures.

Re:Binocular Vision

[Anonymous Coward]

Binocular from 1 camera and mirrors could also work. It would be just like a left-right split-screen.

      If you know the distance between the two "eyes" you can determine the actual distance to the object based on the disparity of the object's occurrence in the left and right images (think Angle-Side-Angle from geometry).

I still would prefer 1 wobbling camera (think how owls move their head sideways). It would provide an "infinity" of perspectives to use throughout the continuous motion instead of 2 fixed perspectives. Correlating which object in the left image corresponds to which object in the right image can be troublesome. With a wobbling camera, each frame is only a slightly different perspective from the previous frame/perspective (not a large discrete jump from left to right perspective) so correlating/tracking is easier.

Re: Binocular Vision

[Anonymous Coward]

TFA gets rid of any plausible deniability. If Volvo ships in Australia with the kangaroo defect and someone hits a kangaroo while in autopilot, they're at fault

Re:Binocular Vision

[Anonymous Coward]

It's the cost of stitching together all those CCD sensor images. Each sensor will have a different overall brightness/contrast/saturation. That has to be compensated for each CCD video buffer (4096x4096 pixels @ 16-bits). Then any fisheye lens effects have to be removed by software/hardware. All the gigabytes of transformed image data then have to overlaid and blended together. Even 360 degree cameras have noticable seams. Some areas won't be of much interest, and some areas will be of extreme interest and may need more magnification or detail, provided by a zoom system

There are military systems that do this like ARGUS or Gorgon Stare:

http://aviationintel.com/skynet-coming-to-a-city-near-you-wide-area-airborne-surveillance-will-be-a-game-changer-over-the-war-zone-and-the-home-front/

Re:Binocular Vision

[Anonymous Coward]

Better yes, but much harder to test.

If you have 3 different systems that all detect objects (Say LIDAR, Camera and IR-Camera) and you make a change to one of them then you only need to test that one.

They will get there eventually.

Re:Binocular Vision

[Carewolf]

All self driving/autonomous/driver assisted cars have minimum 2 cameras.
Facepalm.

They have a lot more than that, but they are usually not mounted in pairs pointing in the same direction.

Re:Binocular Vision

[angel'o'sphere]

They usually always have one pair that point in the same direction.
That was my point.
However I got surprised after I wrote my previous comment as I saw a Hunday with driver assist system afterwards that in fact only had one camera pointing forward instead of a pair.

This just in:

[Anonymous Coward]

Kangaroo/car crashes surge in Australia, authorities baffled.

Re: This just in:

[Anonymous Coward]

Bang goes another kanga on the bonnet of the van.

Re: This just in:

[Anonymous Coward]

+10 for Kate Bush reference. If I had mod points...

Re: This just in:

[MichaelSmith]

Somehow I don't think KB will be doing any more tracks with Rolf Harris.

uhuh, long way to go

[holophrastic]

These are the types of things that just make all of the hype totally obviously completely garbage. No animal on the planet would run into a kangaroo. Perhaps the entire concept of the sensor is just way off.

I have memories of the way machine guns evolved. From large gatling systems to the smallest of springs. These cars are still at the large gatling systems.

Re:uhuh, long way to go

[religionofpeas]

No animal on the planet would run into a kangaroo.

Except for humans. According to TFA there are 16,000 kangaroo strikes per year.

Re:uhuh, long way to go

[OrangeTide]

Not very many 1200 kg animals roll along the ground at 120 km/h. Newton's first law of motion combined with the square-cube law makes this a non-trivial problem. I think the key to self-driving cars is for them to not kill any more people than human driven cars already do, luckily for self-driving cars we human drivers established a fairly easy baseline to pass.

With your analogy, we didn't simply give up, we still used machine guns even if they were not perfected.

Re:uhuh, long way to go

[drinkypoo]

No animal on the planet would run into a kangaroo. Perhaps the entire concept of the sensor is just way off.

Except for humans. According to TFA there are 16,000 kangaroo strikes per year.

Seems like you're both right (although you're much more right.) Humans' sensors aren't great for depth perception at night, which is when most animal strikes occur — and according to WP, most roos are nocturnal. If the excuse is that the roo "looks" like it's closer or further away depending on whether it's in the air, then I suspect the flaw is inadequate depth sensing. The car should be dramatically better at this than a human before it's allowed to run around at speeds over a walking pace.

Re:uhuh, long way to go

[xevioso]

But these are all software problems, and we humans have gotten pretty good at that. It's an interesting case here because it's an unusual animal that travels in a way most other animals don't, but there's nothing here that can't be fixed by adding in some code.

What's more interesting to me is that Volvo is trying to test it's cars in Australia, where there's long stretches of unpaved roads and all sorts of weird things that cross the road. Does the car stop when it sees a snake in the road crossing? How big does the snake have to be before the car decides it will just roll over it? How does it determine if it's a snake or a cable or rope? These are all software things though...

Re:uhuh, long way to go

[religionofpeas]

The good thing is that the car's sensors and algorithms can be fixed. It sounds like it's mostly a case of Volvo overlooking the effect of jumping animals. According to the Volvo information about their autonomous cars, they have camera, radar, laser, and ultrasonic sensors. With the right algorithms, the car should be much better than a human at detecting animals at night, and respond much quicker too.

The car should be dramatically better at this than a human

Better than a human is good enough for now.

Re:uhuh, long way to go

[Wycliffe]

The good thing is that the car's sensors and algorithms can be fixed. It sounds like it's mostly a case of Volvo overlooking the effect of jumping animals. According to the Volvo information about their autonomous cars, they have camera, radar, laser, and ultrasonic sensors. With the right algorithms, the car should be much better than a human at detecting animals at night, and respond much quicker too.

There will always be unknown objects at or near ground level. With camera, radar, laser, and ultrasonic sensors, how did they screw up the depth perception that badly? Will it be able to detect a small child? What about a medium size child? What about a box? What about a bumper that fell of a car? What about a flock of birds? What about a chair,cage of turkeys,sofa,trashcan,bale of hay, etc.. that fell off a truck. If you're designing it for known obstacles, you're likely doing it wrong. Every human driver out there is able to detect something that isn't suppose to be there and take corrective actions even if they don't know what the unknown object is.

Re:uhuh, long way to go

[drinkypoo]

Every human driver out there is able to detect something that isn't suppose to be there and take corrective actions even if they don't know what the unknown object is.

Yeah, this. If I don't know what a thing is, I still hit the brakes, because I don't know what its density is and that might be important. If it turns out to be a plastic bag or something, well, that's OK so long as the person behind me wasn't following too close. And since I have a nice safe car with lots of crumple zones, frankly I don't really give a shit if someone drives up my ass as long as I don't go off the road, and they have enough insurance to replace my nice safe car.

Re:uhuh, long way to go

[GungaDan]

"How does it determine if it's a snake or a cable or rope?"

Or a cop's spike strip...

Re:uhuh, long way to go

[religionofpeas]

Unlike the summary suggests, the Volvo is not a driverless car. It's a car with a Driver Assist function, with a special module to warn for large animals. The first step is to signal the driver, and provide information on the type of object. The driver can then decide what to do. Only if the driver doesn't react on time, the system takes over and brakes. It's still a fairly simple system.

Keep in mind that large wild animals are also potentially dangerous if they are not in the middle of the road, but standing next to the road, thinking about crossing.

Re:uhuh, long way to go

[holophrastic]

...and just look at how many people were killed by bad machine guns. The car needs to be way better than a human before it's kills are acceptable. I can accept another human making a mistake that I would make. I don't accept a machine making a mistake that I would make. I already find it tough enough to accept a human making a mistake that I wouldn't make.

In any event, how's this for an algorithm: look forward. if you see the you expect to see (a road) keep driving. If you see fur, instead of a road, slow down. If it keeps getting bigger, slow down more. If it gets smaller, speed up. Doesn't matter if it's in the air, or on the ground. If a part of it is an eye level, slow down if it's getting bigger.

Who cares what it is, or how it's moving. It's either getting bigger, or getting smaller, as you move forward. That's it.

Re:uhuh, long way to go

[tsqr]

The good thing is that the car's sensors and algorithms can be fixed. It sounds like it's mostly a case of Volvo overlooking the effect of jumping animals. According to the Volvo information about their autonomous cars, they have camera, radar, laser, and ultrasonic sensors. With the right algorithms, the car should be much better than a human at detecting animals at night, and respond much quicker too.

According to TFA, Volvo has been working on this problem for two years. I'm guessing the "right algorithms" have been right around the corner for some time now.

Re:uhuh, long way to go

[tsqr]

Unlike the summary suggests, the Volvo is not a driverless car.

It's not just the summary. The TFA is entitled, "Volvo's driverless cars 'confused' by kangaroos", and contains such suggestive phrases as, "Volvo's self-driving technology is struggling to identify kangaroos in the road" and "the problem would not delay the rollout of driverless cars in the country".

Re:uhuh, long way to go

[Anonymous Coward]

and according to WP, most roos are nocturnal.

No, roos are not nocturnal. I live in the bush and have a friend that rehabilitates rescued Joeys (baby roos) and you don't drive certain roads at dawn or dusk as the roos are on the move seeking water. That's when most of the strikes happen.

Re:uhuh, long way to go

[Anonymous Coward]

In the last 4 days I've watched two broad daylight collisions on a wide open intersection where both vehicles were entirely visible so your assertion about corrective actions is clearly wrong in some cases. If you think that humans run into things less frequently than automated cars, even at this early stage, I'd suggest you are probably wrong.

Re:uhuh, long way to go

[Anonymous Coward]

If it gets smaller, speed up.

Test Engineer #1: Ok, cue the grizzly bear.
Test Engineer #2: See how the sensors detect the bear as it crosses the road and automatically slow down the car?
Test Engineer #1: Perfect. That test passes.
[Bear stops and turns to face the car, presenting a smaller profile]
[Car speeds up and crashes into bear]
Test Engineer #1: Shit.

Re:uhuh, long way to go

[angel'o'sphere]

Volvo is reporting that they have "trouble" to judge the distance.
They did not say they won't stop or have problems to detect the Kangaroo at all.
So your questions about kids etc. are rather absurd.

chair,cage of turkeys,sofa,trashcan,bale of hay, etc
Most of them don't fly ... so you have no problem to judge the distance.
Or is it easy for you you to judge how high a bird is flying which species (aka size) you don't know?

With camera, radar, laser, and ultrasonic sensors
Radar does not pick up Kangaroos. (They are made of meat, you know ...)
Nor does ultra sonics, as ultra sonic sensors are used for close range (e.g parking) and the street surface.

Re:uhuh, long way to go

[angel'o'sphere]

I'm guessing the "right algorithms" have been right around the corner for some time now.
It is most likely not an algorithm problem but a sensor problem. Or kets say a combination. How to interpret the sensor data in a meaningful way.

Re:uhuh, long way to go

[angel'o'sphere]

Doesn't matter if it's in the air, or on the ground.
Of course it matters. Without any "cross reference point" you can not judge how far something is away if it is in the air and the background behind it is only sky.

look forward. if you see the you expect to see (a road) keep driving. If you see fur, instead of a road, slow down.
That is what the car is doing. Volvo only stated that there are troubles to get the distance accurate.

Re:uhuh, long way to go

[fluffernutter]

Unlike many Slashdotter's suggest, there is no self driving. Only cars following rules that are too rigid for real world use.

Re:uhuh, long way to go

[freeze128]

I would not want my self-driving car to take corrective action if it were about to hit a flock of birds. That would more likely than not cause an accident. Birds have flown across the road at low levels in front of my car all the time, and I have never slowed down or swerved, and I have never hit one.

Re:uhuh, long way to go

[ColaMan]

It's not about depth perception.

Roos are erratic creatures and can move very quickly. So one will be grazing 20 metres off the side of the road, become spooked by your headlights, bound onto the road in two or three hops, and hit your car.

It's about a second per hop for roos, so this takes place very quickly, well inside the illuminated area that your headlights project. Often they will appear from the vegetation on the side of the road and then be on the road in one hop, only 10 or so metres in front of you. It's why most cars in rural Australia are festooned with LED light bars and spotlights because the further and - most importantly - the wider you can see at night, the better.

Here are a few typical roo strikes to give you some idea of the problem -

https://www.youtube.com/watch?...
https://www.youtube.com/watch?...
https://www.youtube.com/watch?... - This one in particular.

Re:uhuh, long way to go

[Anonymous Coward]

Kangaroos are actually really hard to avoid. They're fast, pretty well camouflaged against Australian bush, move unpredictably, tend to be active at dusk when visibility is poor, and have precisely zero instincts telling them that jumping out in front of a car is a bad idea. Driving along a rural road with lots of tree/bush cover around sunset is a genuinely risky business.

Re:uhuh, long way to go

[dlingman]

If it gets smaller, speed up.

Test Engineer #1: Ok, cue the grizzly bear.
Test Engineer #2: See how the sensors detect the bear as it crosses the road and automatically slow down the car?
Test Engineer #1: Perfect. That test passes.
[Bear stops and turns to face the car, presenting a smaller profile]

[Car comes to complete halt.]
Test Engineer #2: Told you we fixed that...
[Bear wanders over to car, peels open roof, eats contents, and leaves.]
Test Engineer #1: Shit.

Re:uhuh, long way to go

[Anonymous Coward]

Except for humans.

Fun fact: It's the kangaroos doing the hitting, not the humans. Erratic movement is a defensive mechanism, but is suicidal around highways. Keep mind that roos can jump a dual lane road in one go, so they can go from hiding in the scrub to smashing through your windscreen in less than a second. Or they'll happily bound along the parallel to the road, and the second you try to overtake them they jump into your grill. Locals are on to this and use defensive driving such as slowing down if scrub is near the edge of the road, avoid driving at dusk, not overtaking a roo if there is oncoming traffic, and maintaining constant awareness that collision could occur at any moment and the worst thing you can do if that happens at highway speeds is to attempt to avoid it.

Re:uhuh, long way to go

[jezwel]

'Roos are leaping away from unknown / danger, which at night is their shadow. This is why they leap towards lights and your car.

Re:uhuh, long way to go

[RespekMyAthorati]

Better than a human is good enough for now

For the research phase, sure. But not for actual deployment.

Re:uhuh, long way to go

[Anonymous Coward]

What about kangaroos that deliberately try to kill themselves?

https://www.youtube.com/watch?v=3kTeoO7IGjc

Re:uhuh, long way to go

[OrangeTide]

...and just look at how many people were killed by bad machine guns. The car needs to be way better than a human before it's kills are acceptable. I can accept another human making a mistake that I would make. I don't accept a machine making a mistake that I would make. I already find it tough enough to accept a human making a mistake that I wouldn't make.

I don't agree with your position, and I doubt the legislature of your country will agree with you either. You're certainly free to have your opinions, and refuse to use self-driving vehicles. But I the rest of your community are not likely to put a halt on this because of the concerns you raise, because I believe they are very much a minority concern that doesn't carry much weight philosophically with most people.

Re:uhuh, long way to go

[Anonymous Coward]

But that's not what the system does. They already have an auto brake system to try to stop you if something is in front of the car.

What *this* system does is that it uses sensors to notify the driver that there is a large animal near the road. And it tries to assess the risk of that animal getting in front of the car. So it basically tries to help the driver notice large animals. This is especially good when they can be found with IR Cameras in the dark.

How ever Volvo also has a program to develop driverless cars and it's a safe bet that the same technology that today is in an assist system will one day form the basis of the self driving car.

Re:uhuh, long way to go

[Anonymous Coward]

You are right, and I can't emphasize this enough, but kangaroos are extremely erratic near roadways. I've seen them bounce away from a road and then turn back at the last minute and bounce right back in front of the car from the side. Heck, I once had a deranged kangaroo bounce into the *back* of my car once while I was crawling along a country road trying not to hit anything. They're fucking crazy animals.

Dodging roos is practically impossible for human drivers, I don't see how a self driving car with all the additional uncertainty from the sensor data can possibly hope to do better. The only safe option is to slow the car to a crawl/stop whenever a kangaroo is seen within 10-20m of the roadside...which is not conducive to getting anywhere in Oz, the little bastards are everywhere.

Re:uhuh, long way to go

[mjwx]

With the right algorithms, the car should be much better than a human at detecting animals at night, and respond much quicker too.

Also much better at false positives as we're finding out. Many of these vehicles are unnecessarily slamming on the brakes over nothing. Not only is this dangerous, it's also expensive with increased congestion (and bad for the environment).

The car should be dramatically better at this than a human

Better than a human is good enough for now.

But they aren't. Computers are reactive, humans are predictive. A good driver avoids dangers before they happen. The only reason Google's autonomous car has been successful is because the human driver is predicting issues. Left to it's own devices it drove right in front of a bus.

Autonomous cars as you imagine them will not be a thing for some time now. Proponents of autonomous cars are counting on technology that doesn't exist at the moment and may not exist or be good enough for use within our life times. This technology does not simply need to be "better than a human", it needs to be so safe it can be given, with confidence, to the dumbest and riskiest humans imaginable. Sorry if this bursts some bubbles.

Re:uhuh, long way to go

[Anonymous Coward]

Unlike many Slashdotter's suggest, there is no self driving. Only cars following rules that are too rigid for real world use.

As opposed to humans, of course, who have perfectly flexible and adaptable rules, and therefore are never at fault for any accident ever.

Re:uhuh, long way to go

[drinkypoo]

It's not about depth perception.

Well, no. And also no. TFA explicitly says that the problem is not that the car doesn't have time to react, but that the car is having trouble identifying the roo as not being a bird, and thus is making incorrect assumptions about how it will move.

It's about a second per hop for roos, so this takes place very quickly, well inside the illuminated area that your headlights project.

So what you're saying is that it's well within sensor range? And so they have absolutely no excuse for not detecting it? Right-o!

Re:uhuh, long way to go

[drinkypoo]

I would not want my self-driving car to take corrective action if it were about to hit a flock of birds. That would more likely than not cause an accident. Birds have flown across the road at low levels in front of my car all the time, and I have never slowed down or swerved, and I have never hit one.

Well, birds do the same thing with me, and I have hit them on occasion. And I have had to brake to avoid hitting them. And I have been able to do that fast enough to make a difference sometimes, and not other times.

Your lack of experience does not constitute proof of a negative. Thanks for playing, though.

Re:uhuh, long way to go

[ColaMan]

Sensor range isn't headlight range.

Time to identify an object emerging from the undergrowth 20 metres away with a varying trajectory is minimal at 100km/hr. It's less than a second. Coupled with kangaroos typical behaviours - like running parallel to the road for a bit as you approach before jumping across in front of you - well, it's going to be tough to do with software.

Roo running alongside the road which is an immediate threat? Or person on a bike, which will maintain speed and position as you pass? Hard to figure that out in a couple of seconds unless you're doing full-on machine vision.

Re:uhuh, long way to go

[drinkypoo]

Sensor range isn't headlight range.

What? Of course it is, at least, some of the sensors can clearly see at least as far as the headlights, not least because they are using a camera in the visible spectrum. Talk sense.

Time to identify an object emerging from the undergrowth 20 metres away with a varying trajectory is minimal at 100km/hr. It's less than a second.

No, I don't give half a squashed shit what it is. If it is looking like it's about to get in front of the car, stop. It doesn't matter if it's a plastic bag that might clog my air intakes, or a roo that might destroy the front of my vehicle, I want to avoid hitting it either way.

Roo running alongside the road which is an immediate threat? Or person on a bike, which will maintain speed and position as you pass?

No, bullshit. A person on a bike doesn't emerge from the undergrowth at umpty-ump miles per hour, and if they do, you want to give them some extra space anyway because they might fall off their bike and go under your wheels. It's completely irrelevant what an erratically moving object is, you want to avoid it. Also, you're changing the parameters again. You started with a mystery object well off the road suddenly charging for the road, and then compared that with a cyclist, which doesn't make sudden mystery moves which are anything like that. Make up your mind!

Re:uhuh, long way to go

[brantondaveperson]

But these are all software problems, and we humans have gotten pretty good at that.

That hasn't been my observation of human's track record with creating software at all. Humans are great at hardware problems, but to this day humans are manifestly terrible at creating software.

Re:uhuh, long way to go

[Anonymous Coward]

My god, but you're a fucking idiot.

Re:uhuh, long way to go

[gwjgwj]

Eats, shoots and leaves.

Re:uhuh, long way to go

[LordWabbit2]

My brother was working on the road trains in the outbacks (he had just immigrated) and the first kangaroo he saw the truck ran over it.
Didn't even stop, the trucks have these huge bull bars on them, they are designed to hit kangaroo's and keep going.
If you don't know what a road train is (I didn't) Road Trains

under the EULA we are not at fault just be lucky

[Joe_Dragon]

under the EULA we are not at fault just be lucky that it was not kid / baby as under the EULA we do not need to do any think to help you in court or wave the NDA so we can give out log's / source code / etc to your legal team.

Not Many Roos in Pakistan

[Anonymous Coward]

Or wherever they make those boxes on wheels.

Note to self

[Valley Redneck]

Avoid Volvo driverless cars in Australia, around zoos with bad security.

Re:Note to self

[Anonymous Coward]

You do know they have kangaroos in the wild in Australia, don't you? They aren't just in zoos.

No!

[Anonymous Coward]

The car AI is right! It is the animals that are wrong!

To be fair...

[Xenolith0]

I'm confused by kangaroos too.

The obvious solution is...

[Nutria]

to kill all kangaroos.

(If I were Bender Bending RodrÃguez, the obvious solution would be to kill all humans, but since I'm not Bender Bending RodrÃguez, that's not the solution.)

Imagine them getting attacked by Drop Bears...

[cant_get_a_good_nick]

then they'll really be hurtin'

Backwards Automation

[Anonymous Coward]

They should have outfitted the kangaroos with electronics to avoid the automobiles. Besides riding in a kangaroo's pouch is way more comfortable than a Volvo..

Re:Backwards Automation

[Thud457]

No.
Obviously they should arm the kangaroos with stinger missiles.

Re:Backwards Automation

[Anonymous Coward]

Would you just arm the roos, or also provide them with training? Maybe put them through the, um, SEAL program?

They confuse me too.

[gurps_npc]

What's with the boxing? Who the hell taught them to fight like men? WHY?

And marsupials are just flip-floppers. Birth but still inside the mother's pouch? Come on out already. Are they going for a second birthday?

Re:They confuse me too.

[ghoul]

Marsupials Millenials. Whats it with Mxxxilas never wanting to leave the mother's basement/pouch

The real problem of the 'self driving car':

[Anonymous Coward]

The real problem is the 'so-called' 'AI' is not a real 'AI' (and I and many others wish they'd stop referring to it as such). It only half-assed recognizes things based on what it's 'trained' to recognize. A human being, seeing some animal they've never seen before in their life but that is crossing the road, is going to swerve or stop anyway because the human brain is actually intelligent. These 'deep learning algorithms' are not even as smart as a dog brain. I think the entire approach is doomed; it shows enough promise to get investors to give them money, but it'll never be as good as a human driver because it's not really 'intelligent'.

Re:The real problem of the 'self driving car':

[Anonymous Coward]

A fruit fly has more intelligence. We have been running AI programs in academia for the last 50 years and have not come anywhere close to the brain power of an animal that has a life span of 45 days.

Re:The real problem of the 'self driving car':

[Anonymous Coward]

Driving is simply a task. Like chess, computers will inevitably be better at it than human drivers. It doesn't need to know how to eat or reproduce to accomplish this task.

I'd actually be very interested in your definition of intelligent if you thing fruit flies are.

Re:The real problem of the 'self driving car':

[Anonymous Coward]

Hm, and how good will your AI do in chess game without rules? In a game, where you can in your first move remove the queen?

With your amazing social interaction skills

[presidenteloco]

I would like to invite you to be a valuable and ongoing member of our technology development team. Also, I would like to subscribe to your newsletter.

Re:With your amazing social interaction skills

[Anonymous Coward]

Is this the new slashdot? Being amazed (and annoyed) on people who reads TFA?

Re:With your amazing social interaction skills

[Anonymous Coward]

No, it's the same old Slashdot where unhinged lunatics scream with rage at other commenters and then get modded down and/or made fun of.

Not the first time.

[Anonymous Coward]

http://www.snopes.com/humor/nonsense/kangaroo.asp

An oldie but a goodie.

Comment

[WallyL]

Wait, so everything Down Under is out to kill you, and now they've developed stealth technology?! We're doomed!

To be fair

[Anonymous Coward]

To be fair, kangaroos can be a bit confusing. I wonder how those cars feel about platypuses. That must be quite a shock.

Volvo Cars

[Anonymous Coward]

Volvo Cars are not Swedish they are a Chinese company

Kind of makes me wonder

[Anonymous Coward]

Kind of makes me wonder what other common everyday occurrences their systems have difficulty detecting, that the designers aren't even aware of, nor would have ever suspected to test for.

No thanks. I'm looking forward to never having to set foot in one of those death traps.

Re:Kind of makes me wonder

[Anonymous Coward]

Pedestrian here. I see how people drive and want everyone to switch over to self-driving cars as soon as possible.
There are plenty of things that human drivers have difficulty detecting that they aren't even aware of, nor suspect to look for.
And those are the skilled drivers.
Then we have aggressive dumbfucks that intentionally takes risks with other peoples lives.
On top of that we have a group of drivers that are too busy surfing on their phones that they don't even bother looking up.

If their main problem right now is that their cars can get confused by kangaroos then we need to get those cars out on the street as soon as possible because the human drivers we have now gets confused by straight roads and regular intersections.

Re:Kind of makes me wonder

[Anonymous Coward]

That's fine, so long as you realize that you will never be able to push one on me (not that I have to worry to much about it, because we can't hope for full autonomy in our lifetime, only driver-assist.)

I trust my own judgement. Not some dopey computer programmer's, who is working for a crooked corporation that will try to weasel out of a lawsuit when their product fails.

I'll maintain control, thank you.

Not only them

[nospam007]

I'm confused as well.

Personally I can't do such leaps with an empty purse.

Close up wallaby

[Hognoxious]

Close up wallaby, or distant kangaroo?

kangaroos can be deadly.

[Anonymous Coward]

You'll be driving on a highway at night, a roo will all of a sudden land in the middle of the road and if they dont stop to look at your headlights they'll be gone in one jump. If you happen to hit that roo, imagine hitting a tree at highway speeds, thats the kind of damage you can expect. The best course of action is to stear around behind the roo or stop if either of those is at all possible without killing yourself, and thats how roos are deadly.

Animals

[fluffernutter]

If they can't detect kangaroo, how the heck do they detect rabbits and birds or anything smaller? Or do animals lives only count when they might damage the precious humans?

Note to self...

[hyades1]

Stop riding pogo stick to work.

Do the kangaroos have rocket launchers?

[Anonymous Coward]

http://www.snopes.com/humor/nonsense/kangaroo.asp [snopes.com]

Wombats are more dangerous

[Anonymous Coward]

Wombats are far more dangerous than kangaroos. Hitting a wombat at high speed, it will probably rip both your axles and transmission shaft right off, and take your fuel tank with it, probably causing an explosion. The wombat will almost certainly survive the encounter and waddle off to its next meal of roots, shoots and leaves.

Seriously... driverless vehicles shouldn't exist!

[squash_me_quickly]

There should in every street vehicle be a person why is responsible/liable the operation of the vehicle.
There should be "driver assistance vehicles" where the driver gets...
- collision detection... warns the driver of obstacles, and possible can apply brakes
- lane change detection... if one crosses lanes without indicating one is "suspected" of not being alert
- blink rate detection... this should warn the driver that they are getting tired, and give the driver a short time to find a save place to park
- male hormone detection... prevents the car from driving more that 60mph
- a "screen light detector"... a camera which point towards the driver and shuts the engine off if it detect a phone/tablet/etc.


This should all be mandatory by 2025.

Australian here

[mjwx]

I expect that Volvo did most of the training with animals they find on the road in Sweden. The White Tailed Deer lives in North America.

But Moose live in Sweden. Moose are much larger than deer, but share the same basic body type (large body on relatively long and thin legs). Special caution needs to be taken with moose, since their eyes don't glow in the dark, and they have tendencies to bite. In fact, a moose once bit my sister.

You see, moose or deer are not analagous to Kangaroos. Kangaroos are as fast as deer, as heavy as moose, as tall as a basketballer and not fucking afraid. Also they jump. Because they have a high centre of gravity, if you hit them they go over the bonnet, through the windscreen and into the occupants in quite a way that a deer or moose does not.

Re:Australian here

[Anonymous Coward]

You see, moose or deer are not analagous to Kangaroos. Kangaroos are as fast as deer, as heavy as moose, as tall as a basketballer and not fucking afraid. Also they jump. Because they have a high centre of gravity, if you hit them they go over the bonnet, through the windscreen and into the occupants in quite a way that a deer or moose does not.

Did someone mutate your kangaroos? Moose can get up to 1500 lbs. Kangaroos, meanwhile, only get up to 200 lbs.