Apple Reportedly Seeking LiDAR Sensors With 'Revolutionary Design' for Self-Driving Vehicles

[ArtOfWarfare]

I think the point is to make it better than human drivers though. As I understand it, LiDAR can sometimes pick up things through foliage at the edge of the road, such as a deer, and can sense things through tinted windows such as cars breaking ahead. The brain computes depth data based on stereoscopic vision. AI/CPU power isn't yet at the level of the brain, so LiDAR helps cheat that for human level depth perception and takes it to another level of performance. In low contrast situations, it could also have an advantage, such as on really foggy days, pouring rain, dust storms, near twilight, etc.

I think you're mixing up radar and lidar. Things which are opaque to visible light cameras are also opaque to lidar. Radar uses a different section of the EM spectrum, so can sometimes see through things which are opaque to cameras, IE, foliage, fog, rain, and dust, as you named.

 

[Gasu E.]

I don't know what Apple wants in the car business but this might be a good time to think about buying someone else to move forward more quickly. With the purchase of Tesla, they would provide the financial backing Tesla needs to continue. They would want to remain hands off for a few years to allow the company to continue as is but partner on technology to learn and at some point combine. Shareholders should ask that Tesla focus on quality and on getting full autonomy to the existing fleet sooner than later, they are already delayed to the point some original buyers are ready to trade cars without ever using the feature they paid for. Some of the early buyers paid 2 years ago. In this bracket of cars, many owners are ready for another new car.

Why would anyone want to pay for the privilege of carrying the baggage of the elon of Ego Musk (did I say that right)? Musk, who is really the primary Series A investor, actually paid for the rights to retroactively claim to be co-founder. The man is entirely over-rated as an entrepreneur (he didn't found PayPal either, although he got himself named co-founder as well), and totally bizarre as a manager, making crazed damaging statements to Wall Street, as well as presiding over massive operations and marketing mistakes. He is a great venture capitalist, but seems to be dissatisfied with have a reputation as merely being a brilliant investor. So, why would you want to pay for a company that has been run by him?

 

[gsmornot]

Why would anyone want to pay for the privilege of carrying the baggage of the elon of Ego Musk (did I say that right)? Musk, who is really the primary Series A investor, actually paid for the rights to retroactively claim to be co-founder. The man is entirely over-rated as an entrepreneur (he didn't found PayPal either, although he got himself named co-founder as well), and totally bizarre as a manager, making crazed damaging statements to Wall Street, as well as presiding over massive operations and marketing mistakes. He is a great venture capitalist, but seems to be dissatisfied with have a reputation as merely being a brilliant investor. So, why would you want to pay for a company that has been run by him?

Simply, he will produce the first car that goes end to end without my involvement. I like performance and like to be the driver but I’m ready for 95% of the time the car drives me. Who else is doing that right now. Most of my driving is mundane errand running that I don’t need to be involved in directly.

 

[jakeenzo]

To all the people saying self driving cars are unicorns, obviously the whole of silicon valley disagrees. And i'm more inclined to believe them over a macrumours keyboard warrior.

 

[macduke]

I think you're mixing up radar and lidar. Things which are opaque to visible light cameras are also opaque to lidar. Radar uses a different section of the EM spectrum, so can sometimes see through things which are opaque to cameras, IE, foliage, fog, rain, and dust, as you named.

Perhaps I am, lol.

 

[jeyf]

To all the people saying self driving cars are unicorns, obviously the whole of silicon valley disagrees. And i'm more inclined to believe them over a macrumours keyboard warrior.

so you all keyboard warrior:
do we expect unicorns

go UniCorns

 

[VaeVictis]

And "spectacularly" killing 1.25 million people worldwide every year (more or less)
Yes, I hear you.

Yes, 1.25 million people are killed each year.

However, currently Waymo is the best at self driving and they reported 0.09 disengagements per 1,000 miles driven (report from 2018)

That is about 11,000 miles driven per disengagement.

We don't know if each disengagement would be a crash or a killed person.

But assume say there would be 1 accident per 3 interventions.

So about 33,000 mlles per 1 accident.


For human data (Arizona 2011) 3,433,955 drivers, each driver averages 14,493 miles.

They had 103,423 accidents and 754 fatalities.

That averages about 3,433,955 drivers * 14,493 miles / 754 fatalities = 66,006,484 miles per 1 fatality

or 3,433,955 drivers * 14,493 miles / 103,423 accidents = 481,216 miles per 1 accident.

There are probably smaller not reported accidents, but currently there is no data regarding them.

So human drivers are about 481,216 / 33,000 = 14.58x better than Waymo or saying otherwise there would be about (1,25 million * 14.58) 18 million people killed if everybody was driven by current tech Waymo (there probably would be some C2C systems and so forth and this number would be lower, but anyway).

And let's not forget Waymo is currently number one at self driving, everybody else has much worse numbers.

Plus human fatalities are declining with newer more safe cars!

We are very far from LVL5 autonomy!


Somebody also mentioned we drive with only two cameras, it is kinda true, but we also use our other senses.


Somebody else mentioned Apple should buy Tesla, this is just stupid (no offence).

Apple would pay more than 50 billion USD (probably somewhere around 60-70 billion - price of premium for take over).

For this amount of money, what would they receive?

- Supercharger network

OK, but for a fraction of the amount (needed for the take over) they could build this themselves

- Tesla brand

Apple brand is much bigger

- Gigafactory

OK, but Apple is notorious about not manufacturing there own products (they would much more likely to use Magna Steyr)

- Autopilot technology

TBH, not that great, many other companies are way ahead, plus they only use cameras, not LIDAR

- EV technology

Great but, batteries are made with Panasonic (currently there are some disagreements between Panasonic and Tesla), EV engine is not that much of a new concept, many companies are making them (even boutique manufacturers like Rimac Automobili).

- Car making technology

Not that great, certainly not within Apple quality

Elon Musk and all the employees

- do they really need somebody like that in Apple, culture clash would be a massive problem

Model S, X and 3

- Model 3 is new, S and X need refreshments

Plus Tesla is practically losing money every quarter.

So all in all, they would pay 60-70 billion USD for Supercharger network and Autopilot, IDK to me that is not great of an investment of that much money.

Plus the main reason why Apple will not do this, their biggest acquisition was Beats and that was for 3.2 billion USD, to expect for Apple to throw 60-70 billion into Tesla is just insanity.

 

[ChrisA]

Can produce a 'revolutionary' autonomous vehicle but can't produce a charging mat?

The mat was to be a consumer product and was cost sensitive. No one would buy a $2,000 charger even from Apple. But these cars have no budget. Each one costs hundreds of thousands of dollars, likely over $1M as there are hand build one at a time using very expensive parts.

A Google engineer said that "his" car ranks in the list on the 100 most powerful supercomputers on Earth. The computer burns 2,400 Watts of power and uses custom "TPU" chips. This tech not NOT cheap because it is not mass produced. Hence the > $1M price.

So, back to my point. It is easier to make something work when you don't have to worry about the per unit price of the product.
[doublepost=1555613137][/doublepost]

Yes, 1.25 million people are killed each year.

However, currently Waymo is the best at self driving and they reported 0.09 disengagements per 1,000 miles driven (report from 2018)

That is about 11,000 miles driven per disengagement.

Yes but you need to define "disengagement".

Very few disengagements would have resulted in an accident. The test vehicles are set to have a very low threshold. Why? These cars are not designed for practical transportation they are designed to collect training data, to train the next version of the car's controller.

In fact, just the opposite, Google NEEDS to collect the human's input in many situations. Disengagement is required if the car is to collect data about "what would a human driver do?"

A very good way to drain a self-drive is to let the human drive it full time. In fact, that is the only way to start the process. But if you don't turn over the car to the human now and then you never pick up any input from the human.

I've actually done this is a car simulator. I drive it around the track 2 or 3 laps and use that to train the steering system. Then let it drive and it crashes so I drive it through the hard section of the tack a few more times.

When these cars are sold to the public it will use a different algorithm for disengagement.

 

[jeyf]

the fate of self-drives will be sealed in the courts and with the insurance companies

we moved back into an urban area and all of our travels are short runs. we would NOT consider owning a self-driver.
i would be VERY interested in a small lite weight EV with:
-small EV like the 1990's cars; Honda Beat roadster or Honda ATCT truck
-accessable battery pack you could swap out
-in the wheel drive motor, eliminate any type of drive train

keep it simple

 

[tipoo]

Not sure if it's an approach or a sensor limitation, but doesn't every lidar method self driving car operate only on limited pre-mapped roads, while general vision approaches (Tesla) are meant to work everywhere?

Is that because of the Lidar, or just the approach of the current companies using it?

 

[LewisChapman]

The mat was to be a consumer product and was cost sensitive. No one would buy a $2,000 charger even from Apple. But these cars have no budget. Each one costs hundreds of thousands of dollars, likely over $1M as there are hand build one at a time using very expensive parts.

A Google engineer said that "his" car ranks in the list on the 100 most powerful supercomputers on Earth. The computer burns 2,400 Watts of power and uses custom "TPU" chips. This tech not NOT cheap because it is not mass produced. Hence the > $1M price.

So, back to my point. It is easier to make something work when you don't have to worry about the per unit price of the product.
[doublepost=1555613137][/doublepost]

Yes but you need to define "disengagement".

Very few disengagements would have resulted in an accident. The test vehicles are set to have a very low threshold. Why? These cars are not designed for practical transportation they are designed to collect training data, to train the next version of the car's controller.

In fact, just the opposite, Google NEEDS to collect the human's input in many situations. Disengagement is required if the car is to collect data about "what would a human driver do?"

A very good way to drain a self-drive is to let the human drive it full time. In fact, that is the only way to start the process. But if you don't turn over the car to the human now and then you never pick up any input from the human.

I've actually done this is a car simulator. I drive it around the track 2 or 3 laps and use that to train the steering system. Then let it drive and it crashes so I drive it through the hard section of the tack a few more times.

When these cars are sold to the public it will use a different algorithm for disengagement.

I’ve not heard of the final Apple car being built by hand before? Do you have a source for that as it’s interesting.

I was also not aware of any product being developed without a per unit price being considered - that’s how the correct market is targeted and achieved.

Also people would pay £2000 for a charging mat if it had an Apple logo on it..

 

[mijail]

The premise that LiDAR is necessary to drive a vehicle is so obviously BS. Humans have been driving vehicles well enough for over 100 years with nothing more than two cameras. Humans can drive vehicles remotely with nothing but cameras. Having LiDAR can certainly assist with driving a vehicle, just as radar and sonar can, but it's obviously not required despite every company that isn't Tesla (or Comma AI) insisting that it is.

You are assuming that when a car crash happens in which an AI was involved, people will be as forgiving (or rather as not-caring) as if it was an all-human disaster. Which is clearly not true, even at present.

And you're assuming that the goal is to have the same rate of accidents as with cars driven by meat blobs with 2 cameras. Which I don't know if it's true, but sounds simplistic - meat blobs and AIs do different things. Case in point: when was the last time you heard discussion about the ethics of a meat blob "deciding" to run over someone to save a passenger vs saving someone outside the car? It's as if the poor meat blob is prisoner to the physics of *driving a car* and maybe it had a bad day and it regrets it oh-so-much, but the AI is an all-controlling entity that will be blamed for *anything*. Why, the AI strictly followed traffic rules, so poor humans were surprised when *they* broke the rules and the AI failed to do the same to accomodate!

Mind you, it is a nice thing to consider the implications of AI - but with a bar that high? Yeah, LIDAR might be actually needed. Who would want to jump into a car driven by the AI equivalent of a trained, semi-evolved monkey?

And Tesla (with its many-more-than-2-cameras) has still no real example to show, right? Isn't it a bit early to compare to them?

 

[PurpleDragon]

LiDAR seems a lot like hydrogen fuel cells - it's smoke and mirrors being thrown up by companies that are lagging behind. Make it sound like you're not lagging behind by saying that you're trying to go do something else.

The premise that LiDAR is necessary to drive a vehicle is so obviously BS. Humans have been driving vehicles well enough for over 100 years with nothing more than two cameras. Humans can drive vehicles remotely with nothing but cameras. Having LiDAR can certainly assist with driving a vehicle, just as radar and sonar can, but it's obviously not required despite every company that isn't Tesla (or Comma AI) insisting that it is.

The problem of the statement that Humans can drive vehicle remotely with nothing but two cameras is that there is an extremely important processing module located behind those cameras called "brain" and it is capable to reconstruct 3D scene. However, AI technology is far behind so pure image processing will not be able to reconstruct the scene with the same reliability and precision as Human brain does. That's why LiDARs are needed. As long as AI is perfected to the level of Human brain - then LiDAR would not be needed anymore.

 

[cmaier]

They are unable to produce a Mac Pro... and they are supposed to be the best on that, go figure.

Unwilling is very different than unable.

 

[Mr Lizard]

Wonder if the design will be as revolutionary as their butterfly keyboards.

 

[ArtOfWarfare]

The problem of the statement that Humans can drive vehicle remotely with nothing but two cameras is that there is an extremely important processing module located behind those cameras called "brain" and it is capable to reconstruct 3D scene. However, AI technology is far behind so pure image processing will not be able to reconstruct the scene with the same reliability and precision as Human brain does. That's why LiDARs are needed. As long as AI is perfected to the level of Human brain - then LiDAR would not be needed anymore.

Tesla gave a 3+ hour presentation on their AI on Monday. Waymo (Google's) lead responded by concurring that Tesla's AI is adequate and lidar is not necessary. (The first 40 minutes and last 10 minutes of the video is just looping filler.)