Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Yes, we are discussing the benefits and limitations of computer vision, object detection and path planning.

I am at least.

We are not discussing whether because life is not a computer game, and some outcomes of actions are unknown, that somehow a computer will never be as good as a human at driving.

No we are not I don't know where you get this idea from.

Because obviously as the human is also operating in this same unknown environment, the statement is flawed, and is not really worth discussing.

Humans make sense of the world and have cognitive abilities that are hard to replicate in a computer. Parsing natural language, making sense of images and so on, it's not a flawed statement, it's commonly understood and accepted.

----------

Indeed. I happen to know this from experience. But my argument, if I was to put it in a sentence, is that this drawback is adequately countered by the significant benefits from reaction time, radar sensors, lack of exhaustion, distraction and frustration and dynamic modelling that a computer is capable of.

I'm not saying that a computer doesn't have benefits...
 
Well, I don't think you can under think this.

It's like a computer game, you have to program in all the rules, the computer does not think it just follows rules there is no intelligence as we would call it.

It's all fine as long as laws and people's opinions can change to accept this.
At the moment, when all else fails and we need someone to blame we have the human behind the wheel we can point the blame at.

If we are to take the human out of the equation then we shall have to find something else to blame, or just accept accidents happen and don't blame.

I can't really see the car makers wanting to accept the blame for every time something goes wrong with their computer driven car.
If you had millions of them, with tens of thousands of lines of code, all in different scenarios, all expecting to run perfectly.

Heck, look at Apple and bugs in just something as simple as a phone or a tablet.
If every car was a computer car and they all talked to each other, that's one issue out the way. They you need to allow for weather, snow, ice, rain, floods across the road.
Even right now, I see a person in a car at a side turning, and I look at their face to see if they are looking at me, if they are not looking at me, there is a chance they might pull out, so I prepare myself just in case. Not much, but I'm mentally ready.

It's going to be really interesting to see who is brave enough to do this 1st.
Or course, when I just stand in front of your computer car, it will be interesting what it will do, just sit there all day, as of course it won't run me over. perhaps it may reverse and try and drive around me :)

Your still overthinking this, what about all the robots we already have eh? In factory's or docks where they have small to large self driving and operating vehicles, they have never hit anything.
They have been developing these cars for years, if Apple is making one it will be a long while before they get it on sale. As I said their is so much red tape they have to go through for a car, it's a completely different world to a phone or computer.
And when i look at what a car you can buy right now can do, self driving isn't far off from the experienced manufacturers. Has the self auto correcting steering on a car malfunctioned and crashed? I've never heard of it, what about the radar cruise control messing up? Never heard of it..

What about plane autopilots? A plane can even land itself if you want, not sure about taking off. But planes every day around the world use autopilots, and if their is a crash it seems to always be human error or mechanical failure or terrorism, not a autopilot fault.
 
Of course we don't know but a lot of the comments seem to be centered around what people think is feasible for Apple as a consumer electronics company, not what the rumors themselves seem to be suggesting. If this project was just around bolstering CarPlay I don't think you'd have a VP of iPhone hardware (a mechanical engineer) leading the project. I don't think Tim Cook would be authorizing a team of 1000 to work on it and I don't think you'd have members of Jony Ive's design team meeting with auto executives/engineers/designers.

Maybe Tim Cook's big ambition is to take Apple beyond consumer electronics. Or maybe he believes the auto industry is right for disruption and that cars are slowly becoming computers. Marc Andreessen tweeted if Tesla can do it so can Apple if if wants to. Last year Tim Cook said in an interview he expected iPhone to be the primary revenue driver for Apple for the next 5 years. Perhaps this project is about what comes after that.

Read my comment again. I didn't say it was bolstering Car Play. I said maybe Apple is working on the brains of a car so, like Car Play (not saying it's the same as Car Play, just easy to install), it can be easily installed in a future model from one of the big manufacturers to get a totally "smart car".
 
No we are not I don't know where you get this idea from.

The original quote:

piggie said:
It's like a computer game, you have to program in all the rules, the computer does not think it just follows rules there is no intelligence as we would call it.

Your interpretation of the quote:

subsonix said:
Lets go back to the original quote, it was said that all rules can be given, just like a computer game. In a computer game all possible outcomes is known and the virtual world has constraints and is described in a non-messy way that is fully understood, it's virtual.

Do I still need to explain where I "get this idea from"?

--------------

Humans make sense of the world and have cognitive abilities that are hard to replicate in a computer. Parsing natural language, making sense of images and so on, it's not a flawed statement, it's commonly understood and accepted.

Certainly. That doesn't mean that we can't in some cases skip the hard stuff, and replace it with an alternate, much simpler algorithm that results in the same, or similar outcomes and performance. The example being replacing exceedingly complex computer vision detection and classification of cats vs dogs vs anything else that the vehicle should avoid, with a radar range detector, that can quickly and easily detect that there is SOMETHING on the road, and give an approximate size. The critical difference is determining what information is needed to make a decision, and what in most cases is irrelevant.

If there's something big on the road, the car shouldn't hit it. Determining if that object is a cat, or a dog, or a child, or a rock, or a 40kg frog, is not necessary.
 
Last edited:
Do I still need to explain where I "get this idea from"?

Yes, the computer game was given as an example, I made a remark how it's different, it doesn't imply that a self-driving car will never be good enough.

Certainly. That doesn't mean that we can't in some cases skip the hard stuff, and replace it with an alternate, much simpler algorithm that results in the same, or similar outcomes and performance. The example being replacing exceedingly complex computer vision detection and classification of cats vs dogs vs anything else that the vehicle should avoid, with a radar range detector, that can quickly and easily detect that there is SOMETHING on the road, and give an approximate size. The critical difference is determining what information is needed to make a decision, and what in most cases is irrelevant.

If there's something big on the road, the car shouldn't hit it. Determining if that object is a cat, or a dog, or a child, or a rock, or a 40kg frog, is not necessary.

Self driving cars usually use computer vision, but regardless, the hard part is to make sense of the world. It's necessary to distinguish between a lamp post and a right exit.
 
Yes, the computer game was given as an example, I made a remark how it's different, it doesn't imply that a self-driving car will never be good enough.

I think it does. It is also a flawed argument, and did not feel the need to discuss it. Yet here we are discussing it.

Self driving cars usually use computer vision, but regardless, the hard part is to make sense of the world. It's necessary to distinguish between a lamp post and a right exit.

A laser range finder can determine where a lamp post is. So can computer vision optical flow. Identifying that it is a lamp post, as opposed to some random solid object, is not necessary. In general, right exits do not have large cylinders in the middle of them. And if they do, you probably wouldn't be using them anyway.

As for identifying right exits, detecting white lines on a road is easy enough using computer vision. And the car can always use an odometer, GPS, 6-axis IMU and magnetometer to estimate where it is, in relation to right exits with the help of a road database.
 
I think it does. It is also a flawed argument, and did not feel the need to discuss it. Yet here we are discussing it.

That's how the discussion started, that doesn't make it a discussion about computer games, I hope you can see that. This is a meta discussion at this point, not a discussion about computer games. My point was never that a self-driving car won't ever be good enough. That's a complete strawman, and further more calling arguments flawed then avoid discussing them doesn't prove anything.


A laser range finder can determine where a lamp post is. So can computer vision optical flow. Identifying that it is a lamp post, as opposed to some random solid object, is not necessary. In general, right exits do not have large cylinders in the middle of them.

As for identifying right exits, detecting white lines on a road is easy enough using computer vision. And the car can always use an odometer, GPS, 6-axis IMU and magnetometer to estimate where it is, in relation to right exits with the help of a road database.

It was an example where the distinction matters, as opposed to your previous examples. I'm aware that computer vision isn't used alone. The fact is that it is a hard problem, which is why it has been an active research field for a long time, your attempts at trying to trivialize it doesn't change that at all.
 
That's how the discussion started, that doesn't make it a discussion about computer games, I hope you can see that. This is a meta discussion at this point, not a discussion about computer games. My point was never that a self-driving car won't ever be good enough. That's a complete strawman, and further more calling arguments flawed then avoid discussing them doesn't prove anything.

You brought it up. I ignored the computer game argument because there was no point arguing it. You then brought it up, and I gave you my opinion on it. Now you won't let it go. Your argument seems to keep changing, and if it's not that a self-driving car is too difficult to make, then I don't know what it is.

I have already outlined why I think the computer game argument is flawed.

It was an example where the distinction matters, as opposed to your previous examples. I'm aware that computer vision isn't used alone. The fact is that it is a hard problem, which is why it has been an active research field for a long time, your attempts at trying to trivialize it doesn't change that at all.

It was an example that can still be solved by bypassing the complex object recognition and classification step of computer vision with much easier, but still functional, alternatives.
 
You brought it up. I ignored the computer game argument because there was no point arguing it. You then brought it up, and I gave you my opinion on it. Now you won't let it go. Your argument seems to keep changing, and if it's not that a self-driving car is too difficult to make, then I don't know what it is.

My argument is that it's not comparable to a computer game at all, on this we seem to agree. My point is that it's not as easy to make a self-driving car as some people suggest, and some level of skepticism about it is warranted.

It was an example that can still be solved by bypassing the complex object recognition and classification step of computer vision with much easier, but still functional, alternatives.

It's one example, but there are an almost infinite amount of combinations of objects, road and weather conditions that can occur.
 
Last edited:
The biggest problem is not, price or computing power, but to give a computer cognitive abilities. It's a research field closely related to artificial intelligence.



It's a different kind of nut to crack, let me put it that way.



Have you tried to reason with a computer? Computers are particularly good at computations and dealing with large amounts of data, given that they are explicitly told what to do. The problem with AI, or any task where the computer is asked to learn, adapt and make judgement on the environment by itself are hard problems.

Why on earth do you think this has ANYTHING to do with AI. You don't need AI for this at all. There is no judgement, ethics, consciousness. It doesn't have to imitate general breadth of computing a human can do. This is a very specialized task.

Do plane Avionics (which are insanely complex need AI) NOT AT ALL.

BTW, my field is systems, computer engineers and I don't see this problem as an AI problem at all. You get info from you hundreds of sensors and you process it to keep your various systems within the parameters defined to keep you away from an accident.

The environment when you drive the streets is a lot more complex than say driving on tracks, but so what. Pour money into it and it will be solved eventually and soon.
 
My argument is that it's not comparable to a computer game at all, on this we seem to agree. My point is that it's not as easy to make a self-driving car as some people suggest, and some level of skepticism about it is warranted.



It's one example, but there are an almost infinite amount of combinations of objects, road and weather conditions that can occur.

While there is a infinite combination. You don't need to care for those infinite combinations. You prune the input to what's actually meaningful to the task, just like a human would do. That's how algorythms work. Not every single things actually matter to keeping your car on the road and not hitting someone. Actually very few things matter. That'S why people are able to drive cars in a quasi zombie state and not kill someone (most of the time).

For example, weather conditions mean nothing if it doesn't impact traction (rain, ice, snow), the forces on the side of the cars (from wind) of visibility (going to the cameras or Lidars) VS you location (a curb versus a straight line) and speed (slow versus versy fast).

A car has a driving envelope, just like a flight envelope; the ride controller could tailor its inputs so it stays inside the capabilities of the cars for the current situation.

Adding random other cars and pedestrians does complicate things. If cars could communicate with each other, then most cars on cars accident would be avoided without even needing to parse the environment. Eventually that will be the case.

But, not all accidents could be avoided. Some things are very unpredictable like hitting a small patch of black ice that sends you in the other lane.

The last hardest issue would be pedestrians and cyclists. Unless they have transmitters with geotrackers on them (then it would be easy). If they don't. Cars would have to scan the environment to avoid their random actions. Takes more computation, but again nothing really insurmountable.
 
I'd love to see what a Jony Ive designed car would look like.

I'm picturing a stainless steel exterior like a Delorean combined with an organic symmetrical shape like a VW Beatle, but more elegant.
 
iCar, twice as expensive as a similar car, and if you don't buy a new one every other year it will drive slower and slower until it eventually just stops :p
 
Does anyone worry that someone will be able to just hack into one of these self-driving cars and tell it drive off a cliff or head-on into traffic or something? There's that idea of safety systems in cars being able to talk to each other over Bluetooth or something similar. Again, you could have people hack into these systems and give false data to cause crashes. With the prevalence of cyber terrorism and bank robberies, etc. these days through supposedly "secure" systems, I have exactly ZERO faith in these "safety" systems being secure enough to trust them on the road, let alone the technology itself.

No one wants to die because the computer decided to brake instead of swerve when an oncoming car crossover over the line. At least when a person makes a mistake, it's their own fault, not some programmer. Thanks, but I'd like to keep driving myself. Plus driving is supposed to be FUN and something to look forward to. If I wanted to ride the bus and text the entire time, I'd go to the bus station.
 
Does anyone worry that someone will be able to just hack into one of these self-driving cars and tell it drive off a cliff or head-on into traffic or something? There's that idea of safety systems in cars being able to talk to each other over Bluetooth or something similar. Again, you could have people hack into these systems and give false data to cause crashes. With the prevalence of cyber terrorism and bank robberies, etc. these days through supposedly "secure" systems, I have exactly ZERO faith in these "safety" systems being secure enough to trust them on the road, let alone the technology itself.

No one wants to die because the computer decided to brake instead of swerve when an oncoming car crossover over the line. At least when a person makes a mistake, it's their own fault, not some programmer. Thanks, but I'd like to keep driving myself. Plus driving is supposed to be FUN and something to look forward to. If I wanted to ride the bus and text the entire time, I'd go to the bus station.

The fact that I hyperventilate at times because driving is stressful to me, which prevented me from ever getting a license, kind of defeats the idea that it's fun. ;)
 
Does anyone worry that someone will be able to just hack into one of these self-driving cars and tell it drive off a cliff or head-on into traffic or something? There's that idea of safety systems in cars being able to talk to each other over Bluetooth or something similar. Again, you could have people hack into these systems and give false data to cause crashes. With the prevalence of cyber terrorism and bank robberies, etc. these days through supposedly "secure" systems, I have exactly ZERO faith in these "safety" systems being secure enough to trust them on the road, let alone the technology itself.

No one wants to die because the computer decided to brake instead of swerve when an oncoming car crossover over the line. At least when a person makes a mistake, it's their own fault, not some programmer. Thanks, but I'd like to keep driving myself. Plus driving is supposed to be FUN and something to look forward to. If I wanted to ride the bus and text the entire time, I'd go to the bus station.

Planes avionics already handled whole plane loads of people in an automated way.

Nobody wants to die period. But,, many still do because we are as a whole, crappy drivers. Our brain doesn't multitask, it does task switching, which means we easily miss things.

Nearly 100% of hacks right now in the world are created through social engineering, not direct attacks. If they isolate the part users can mess with by putting some crap on it (say the entertainment system) from the driving systems, everyone will be just fine.
 
  • Like
Reactions: radiologyman
I don't think Apple is building their own car--the engineering involved is enormously expensive, to say the least.

A far more likely scenario is Apple working with the like of BMW or Mercedes-Benz (both of whom have research offices in Silicon Valley) on building a technology demonstrator using Apple technologies based on something like the BMW 2-Series Active Tourer or the Mercedes-Benz B-Class hatchback.
 
  • Like
Reactions: radiologyman
Artificial intelligence would be one way to do it but technology may not get there in a while.

Breaking down driving to a finite number of tasks that computer can handle will happen sooner rather than later. And soon after computer will be a safer driver given how prone humans to make errors. You can analyze error made by a single computer and reprogram all other computers to avoid repeating it. Not the same for humans.
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.