Apple Shares Research into Self-Driving Car Software That Improves Obstacle Detection

[ceriess]

I am not in the automotive business, but as a car manufacturer I would hardly get anxious (or excited) about Apple and its selfdriving whereabouts.
The metadescription of this publication doesn't fit the picture (huge camera's) - most car companies have more compact solutions than Apple that have proven themselves on the road by now.
My 7 year old Volvo (adapt. cruise ctrl) can discern a pedestrian from a bike or car.
I really do not understand what Apple wants in this arena. No company wants to partner with them. They don't seem to get anywhere themselves (until they buy Tesla...?)
I for one would avoid Apple software - in its current state - in any mission critical system.

Yes, you for one can hold that apparently butt-hurt opinion. Others do not have to follow your lead. I wonder how rich you are with your spot on business advice?! LOL...

 

[robinw77]

Does „cloud based“ mean that it will recognize obstacles as reliably as Siri understands my requests and not work at all when there‘s no reception?

It doesn't even say Cloud Based. It says POINT Cloud Based, which is something completely different. It's pretty clear if you click through to the actual article as well.

From the article:
The paper is titled "End-to-End Learning for Point Cloud Based 3D Object Detection"

 

[Taipan]

It doesn't even say Cloud Based. It says POINT Cloud Based, which is something completely different. It's pretty clear if you click through to the actual article as well.

From the article:
The paper is titled "End-to-End Learning for Point Cloud Based 3D Object Detection"

Oops, sorry!

 

[robinw77]

Oops, sorry!

No worries, glad to help!

p.s. I've just read my original reply back and apologies if it comes across as condescending. I hadn't had my second cup of morning tea at that point!

 

[bollman]

I've said it before, and I'll say it again:
There has to be money made.
I firmly believe that the first mode of transportation that will be "self driving" is airplanes. There's lot of money to be saved for the financially pressed airlines, they are sure willing to pay.

 

[BlargKing]

Even if self-driving cars were barely half as good at humans at dealing with all the stupid driving mistakes made by humans, if we got rid of all the human-driven cars on the freeways, they would immediately be orders of magnitude safer than humans, because they wouldn't have to deal with any real surprises, and would have much, much better reaction time when dealing with the car ahead of them slowing down or speeding up. This would mean a much faster average speed on freeways, because:

• They can maintain a much shorter distance to the next car safely.
• Their faster reaction time would nearly eliminate the accordion action that contributes so heavily to traffic backups.
• Their multiple cameras allow them to merge more accurately, causing fewer delays.
• They don't fall asleep at the wheel.
• They don't exceed the maximum safe speed for the road.
• They don't aggressively, angrily try to pass other vehicles and cause worse backups behind them when they cut people off (and sometimes get hit).

So I want a self-driving car yesterday. Nearly all car accidents are directly attributable to human error, and I have little faith in humans ever being able to reliably do something so dynamic with only two eyes that can only point in one direction at a time. They're hopelessly crippled compared with machines. I hope there is a law within the next six months requiring all cars to have cameras and sensors built-in to make future self-driving upgrades possible, and I hope that human-driven cars are restricted to under-35-MPH city streets within the next ten years, if not sooner.

Yeah they aren't going to force cameras and sensors into cars that soon. It costs far too much.

And you can be sure if ever I get a car chock full of sensors that flash me with warnings Ill be ripping those sensors clean out and driving just fine without them like I've done the past 6 years.
[doublepost=1511460381][/doublepost]

It will likely be decades before self driving cars are in the majority. But I bet soon as that happens people will pass laws to keep human driven car off major streets and urban areas.

People will be terrified at the idea of human driven cars. And they should be. Today a human driven car is the NUMBER ONE CAUSE OF DEATH for people under the age of about 40.

I think it is reasonable to bad something that is the #1 killer.

Self driving cars don't need to be perfect, just good enough that they are say the #2 cause of death would be a huge improvement.. World wide about 1,000,000 people die in car accidents. IN the US the number is 20,000. Think of the World Trade Center attack on 9/11. Cars kill as many people as 10 of those attacks happening every year. Of course they will ban human driven cars just as soon as such a ban is practical.

I guess my driving days are numbered then. I refuse to hand the wheel over to a computer.

 

[AxoNeuron]

I work in the machine learning field and this is ridiculously impressive.

The sheer computational power required to train an end-to-end algorithm like this must have been incredible. Probably required a supercomputer worth tens of millions of dollars.

End-to-end algorithms like this generally require enormous amounts of data, too. I would have hated to have been one of the people running through thousands of hours of LIDAR recordings to try and label where the pedestrians are, must have been exhausting work.

 

[Bacillus]

Yes, you for one can hold that apparently butt-hurt opinion. Others do not have to follow your lead. I wonder how rich you are with your spot on business advice?! LOL...

Thx for your concern.
I am rich enough (whatever that may contribute to...showoff pranks with golden watches?) to have 3 cars, where the oldest is more sophisticated than what Apple has ever done with cars. And I will happily avoid any mission critical stuff from a company whose main OS seems to be in a permanent beta stage.
Happy hiking getting anywhere with Siri (my voice-steered Volvo navi actually understands my commands)

 

[Piggie]

Or perhaps it would not be a better option. The best thing about an autonomous car would be it may have detected that child very early and slowed down to avoid the conflict.

Car vision/sensor system are far superior to humans. Seeing in darker conditions than we can, able to detect and classify objects at much farther distances, and in 360 degrees at once. Then this information can be used to determine how to avoid a conflict.

In an ideal world yes, but we don't live in one.
Car driving at say 40mph and you of a child walk out from behind a van/bus, instantly in front of the driverless car.
It see's your shape, identifies you as an object in it's path. Perhaps can tell you are a person and not a tree branch of something else, then it will have to either hit you as there is not enough stopping distance, or swerve to the right.

It can't swerve to the left as there are parked cars/vans so no room, only direction is to the right, but there is a car coming in the other direction.
So some calculation of, if I swerve to the right to avoid the human and into the path of the incoming car, will that car have enough time to stop when it see's me veer into it's path.
A head on collision possibly? That's you and THAT driver potentially injured.
Perhaps it's best to just brake, and hit the child? at 12 mph and avoid you the car's passenger or someone else being hurt.
And all this must be worked out in a split second.

News Story later that night. "Driverless car runs over & kills child, did not attempt to steer to avoid"

 

[jerryk]

I work in the machine learning field and this is ridiculously impressive.

The sheer computational power required to train an end-to-end algorithm like this must have been incredible. Probably required a supercomputer worth tens of millions of dollars.

End-to-end algorithms like this generally require enormous amounts of data, too. I would have hated to have been one of the people running through thousands of hours of LIDAR recordings to try and label where the pedestrians are, must have been exhausting work.

That is what interns and grad students are for.

Also, since one can use Transfer Learning and other techniques, the training may not be too labor intensive. Thank goodness that the community open-sources models such as Inception.
[doublepost=1511533409][/doublepost]

News Story later that night. "Driverless car runs over & kills child, did not attempt to steer to avoid"

This is going to happen. And people will get upset and focus on the one incident, instead of the thousands of times the automate system prevented an accident.

But in the broader context, the fact remains that 35,000+ die each year in traffic accidents in the US, and 1.2 million worldwide. And over 90% of those deaths are due to human error.