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

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:Me too.

[TechyImmigrant]

I'm confused by Kangaroos too.

Let's not jump to conclusions.

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: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

[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.