Why Tesla will win the Autonomous Driving race

[green48]

Not sure if this has it's own thread, wanted to get your opinions on how Tesla will beat GM, Ford, Mercedes, Google, Uber, etc... in autonomous driving.

I think it's fair to say Tesla has the most advanced product on the market today, it's also probably fair to say Tesla has the best talent, though it definitely doesn't have the deepest pockets in the industry.

What factors will differentiate these products? How do you see this race to driverless cars playing out? Is it just going to be a talent war? Will there be more than 1-2 winners?

Discuss.

 

[Fallenone]

The single most deciding factor is data. With the ever growing fleet of Tesla, all other potential competitors combined can't rival the amount of Tesla has and the gap is not decreasing. Everyday, Tesla increases the size of the database equals to what Google collected over 7 years.

MBLY is partnering with other OEMs and collecting data to build a high precision map too. But the richness of this data is still not comparable to Tesla's. So in all likelihood, MBLY is more likely to create a high precision map first. But that's one of the many pieces of autonomous driving. In the domain of richness and amount of real-world data, no one comes close to Tesla at all. Autonomous driving development is also more and more relying on machine learning. And the key to machine learning is data. The one who doesn't have the best data cannot possibly become the best player. This is a component that neither talent nor money can grant you overnight.

 

[MitchJi]

OTA updates.

 

[schonelucht]

The single most deciding factor is data.

Why?

Also

What data exactly is Tesla gathering? How does it help making autonomous driving possible?

 

[Drivin]

Happy to Discuss.

When is the race being held?
Depends on if it is a long distance race vs. a drag race.
For instance, if the race is 450 miles, then an ICE will do probably do better since it doesn't lose time recharging so even if less efficient in general, overall it can do better.

Discuss.

 

[Fallenone]

Why?

Also

What data exactly is Tesla gathering? How does it help making autonomous driving possible?

I think I sort of answered your "why" question at the end of my post. Autonomous driving at its current state is more like a research question. One of the critical approach to solve it is machine learning (George Hotz is very vocal about this, we don't need to agree with him completely, but there's merit here). And the core of machine learning is to train the machine with data, the more data you have, the better the machine becomes. So data is the core of autonomous driving IMO.

As I don't work for Tesla in autonomous driving, I can't possibly know what they are gathering. But MBLY only has the data from the camera, they may have more with their partnership with other OEMs but unless the other OEMs have the fleet hardware equipped as Tesla AND willing to share ALL the data they collect with MBLY, MBLY still won't have the complete picture of what's going on with the fleet and thereby not having the best data. Plus, I'm not aware of any OEM can have their cars upload data collected easily. I think they would either build a dedicated fleet for research purposes like Google, or asks owners to check in with their representatives in order to collect any data logged. The dedicated fleet can't match the size of data Tesla fleet. The asking owners to check in to give them data just won't work. OTA benefits Tesla owners and Tesla at the same time.

 

[techmaven]

It is interesting to hear George Hotz's take on autonomous driving:

I'm saying he's right, but he is someone deep in the research and development to actually try to make this work.

Also, if Tesla's Autopilot was merely a re-packaging of MobileEye, why would a more advanced MobileEye implementation (2017 Mercedes E-class DrivePilot) be so much worse in the real world? Here's Alex Roy's take:

The War For Autonomous Driving: 2017 Mercedes-Benz E-Class VS. 2017 Tesla Model S
and his follow up:
The War For Autonomous Driving, Part Deux: The 2017 Mercedes-Benz E-Class

And here's another comparison:
Hands off

 

[aesculus]

Yeah. At this point where they are all shooting in the dark on methods, algorithms and models, machine learning and big data is what's going to differentiate the players. So the question is who has the best infrastructure and data scientists to do that along with sufficient data sources to make the domain relevant. I think the answer is singular.

 

[schonelucht]

I think I sort of answered your "why" question at the end of my post. Autonomous driving at its current state is more like a research question. One of the critical approach to solve it is machine learning (George Hotz is very vocal about this, we don't need to agree with him completely, but there's merit here). And the core of machine learning is to train the machine with data, the more data you have, the better the machine becomes. So data is the core of autonomous driving IMO.

But data itself is worthless. What matters is the information contained inside that data. And here is the kicker : Tesla's sensors are simply not advanced enough to gather the data needed for autonomous driving, it's data that is merely good enough to assist a driver.

Even worse : people have monitored the data being sent from a Tesla, @wk057 being the most well known here. It's basically too little to even contain the full information captured from a single camera. Worse yet, we know that Tesla is also sending back mundane information like GPS coordinates (every single manufacturer has multitude of orders more of that GPS coordinates than Tesla, TomTom likely even several multitudes of orders) and things like CANbus data on throttle, braking positions, if the radio is streaming, etc that are no help at all for autonomous driving. So from the meagre stream of data sent back, we also have to deduct whatever is taken up by data that is useless to feed in the machine learning algos.

To recap, not only is the sensor information gathered by a moving Tesla not enough to aid autonomous driving, 99% of what is captured is thrown away. Put this way, a billion miles of data is suddenly reduced to only 10 million miles of information.

 

[roblab]

In my opinion, what you're missing is the "want to". Tesla is known to be a young, motivated, smart, forward looking company. Their competitors are mostly old, outdated, wanting to stay where they are. Even if they HAD a car that could compete, they don't really care.

And then there's the electric car vs. the hundred year old technology at MB, BMW, GM etc. etc. Even though some have electrics, that platform is not being pushed, just like AP is not being developed with much seriousness. It's like GM back 40 years ago. The electric group wanted to move into the future, but the accountants and managers needed to stay where they were. And are.

 

[schonelucht]

In my opinion, what you're missing is the "want to".

Just to be clear. I am not arguing that Tesla does not have a shot at making the best autonomous driving car. I am arguing that the data they are gathering now is not such a big advantage as is argued here. I actually agree them being young and having a silicon valley mentality to think outside the box and radically start over when you reach a technical barrier (see dumping MobilEye) is probably a much bigger plus.

 

[Fallenone]

But data itself is worthless. What matters is the information contained inside that data. And here is the kicker : Tesla's sensors are simply not advanced enough to gather the data needed for autonomous driving, it's data that is merely good enough to assist a driver.

Even worse : people have monitored the data being sent from a Tesla, @wk057 being the most well known here. It's basically too little to even contain the full information captured from a single camera. Worse yet, we know that Tesla is also sending back mundane information like GPS coordinates (every single manufacturer has multitude of orders more of that GPS coordinates than Tesla, TomTom likely even several multitudes of orders) and things like CANbus data on throttle, braking positions, if the radio is streaming, etc that are no help at all for autonomous driving. So from the meagre stream of data sent back, we also have to deduct whatever is taken up by data that is useless to feed in the machine learning algos.

To recap, not only is the sensor information gathered by a moving Tesla not enough to aid autonomous driving, 99% of what is captured is thrown away. Put this way, a billion miles of data is suddenly reduced to only 10 million miles of information.

MBLY is collecting 10 kb of data every km driven. In your opinion, how much data in it is useful information for autonomous driving? And I have to say I totally disagree of you viewing the throttle, braking position, radio status are not helpful. These are data useful to know how the car was being operated. Autonomous driving is not just about the surroundings of the car, its also about the car itself - how it is interacting with its surroundings. And even these information is totally useless, it doesn't discount the 1 billion miles to only 10 million miles useful. They still have 1 billion miles of useful data, along with many other 1 billion miles of "useless" data. This data deposit itself is invaluable. The reason why you need a lot of data is because accidents are rather rare events. Without sufficient data, and we're talking about billions of miles here, you cannot gather enough "system failure" cases to construct a valid statistical analysis to improve your system in order to reduce or even eliminate future system failures.

 

[Phil Seastrand]

Maybe we should define what the finish line is before we argue who is winning.

 

[schonelucht]

Finish line is simple : a car that is able to operate (driving) without human assistance (at the wheel or remote) as a taxi in the larger Bay area. That means the car must be practical for passengers to use, must be able to run at speeds normal taxis run at and must be able to navigate to a passenger without anyone in the car or doing active steering at a remote control station.

 

[Saghost]

Why?

Also

What data exactly is Tesla gathering? How does it help making autonomous driving possible?

The other thing about more data/more cars driving more miles that I didn't see mentioned is the corner cases.

By having a lot of cats driving a lot of miles with active sensors (whether the AP is actually driving or not,) Tesla is much more likely to come across the rare events and these places where the current algorithm is wrong - and if as has often been speculated the car is running AP steering/throttle calculations in the background and comparing them with driver performance, they'll understand the edges of the system and the places it doesn't work well much better than anyone else can without having tried it out for millions of random miles.

 

[tftf]

I think it's fair to say Tesla has the most advanced product on the market today, it's also probably fair to say Tesla has the best talent,
Discuss.

Tesla is the best as you mentioned (best products and best talent, no question). They will win this race, nobody can catch up. Everybody (car industry and new entrants) will have to license Tesla's superior system in the long run. Tesla will be the gatekeeper.

Tesla will also build its own Uber/Lyft competitor to showcase their technology. Additional licensing fees and another huge market. Uber/Lyft again will have to license Tesla's superior system or go under. That's why "Tesla Mobility" will also dominate globally.

Together with the massive SCTY synergy effects in energy storage it's becoming crystal clear why TSLA will become a trillion-dollar market cap company within a decade!

No brainer (as Elon said).

 

[aesculus]

By having a lot of cats driving a lot of miles with active sensors

This did not work out so well ...

 

[Saghost]

This did not work out so well ...

I meant a lot of cars driving a lot of miles - my Swype keyboard on the phone seems to really like cats for some reason. That's a cute image, though.

 

[Yuri_G]

Tesla is the best as you mentioned (best products and best talent, no question). They will win this race, nobody can catch up. Everybody (car industry and new entrants) will have to license Tesla's superior system in the long run. Tesla will be the gatekeeper.

Tesla will also build its own Uber/Lyft competitor to showcase their technology. Additional licensing fees and another huge market. Uber/Lyft again will have to license Tesla's superior system or go under. That's why "Tesla Mobility" will also dominate globally.

Together with the massive SCTY synergy effects in energy storage it's becoming crystal clear why TSLA will become a trillion-dollar market cap company within a decade!

No brainer (as Elon said).

Glad you added a disclaimer to your posts. I'm going to post it here for posterity and to have a nice laugh at in the years to come.

Tesla's planned 2018 ramp from niche producer to 500k cars a year and a mass-market Model3 at $35k base is a financial suicide mission. The proposed SCTY merger is the newest implosion catalyst. Your opinion may differ.

 

[CarlK]

It's definitely the team. The reason all tech companies want to be in the Silicon Valley is that's the place where talents are and where you can build a really strong team from top down. Elon knows, and can tell, who are those best hardware, software and AI people and he will put every effort to seek them out. Those top brains also want to work for a company that their contribution makes the most difference and the top guy speaks their language. Whereas working for an old time auto company you will be a second class citizen to those "car" people who got key positions and are making decisions not entirely based on technical merits they don't necessarily understand. You ended up getting some mediocre people who are better at making powerpoint presentation than real technology breakthrough.