Another company is dialing back expectations for self-driving taxis

[AbsolutSurgen]

Another company is dialing back expectations for self-driving taxis

Quote

Daimler is planning to "rightsize" its spending on self-driving taxis, Chairman Ola Källenius said on Thursday. Getting self-driving cars to operate safely in complex urban environments has proved more challenging than people expected a few years ago, he admitted.

 

Quote

He is just the latest executive to acknowledge that work on self-driving taxi technology is not progressing as fast as optimists expected two or three years ago. Earlier this year, Ford CEO Jim Hackett sought to dampen expectations for Ford's own self-driving vehicles. Industry leaders Waymo and GM's Cruise missed self-imposed deadlines to launch driverless commercial taxi services in 2018 and 2019, respectively.

 

 

Quote

The disappointments of the last couple of years suggest that the most ambitious projects on this list may be biting off more than any company can chew. Tesla's self-driving progress has been much slower than CEO Elon Musk predicted three years ago. Waymo and Cruise have found that it's quite difficult to develop a metro-wide driverless taxi service—even if you have the vast resources of Alphabet or GM at your disposal.

 

The media is finally starting to understand that the hype about self-driving cars was largely made by executives who were trying to prove their companies were tech-savy, and forward looking, rather than any real progress that had been made.

 

When it does come to market, it is going to be VERY expensive, data and computationally demanding, and primarily used for commercial applications for the first decade (or so).

 

 

[legend]

"...has proved more challenging than people expected a few years ago"

 

More than *some* people expected. I certainly didn't expect it at the timescale these companies have been claiming and most other AI researchers with whom I interact we're not as optimistic either. 

[Ghost_MH]

Just now, legend said:

"...has proved more challenging than people expected a few years ago"

 

More than *some* people expected. I certainly didn't expect it at the timescale these companies have been claiming and most other AI researchers with whom I interact we're not as optimistic either. 

 

By *some* people, let's just assume we're talking about rich execs that thought this would be some cheap R&D spent over a few years.

[b_m_b_m_b_m]

But I was told by well dressed people that we should plan the entire future of our cities, from transit to zoning to walkability, based on the idea that we will have driverless cars by 2020

[TwinIon]

While I think it's clear that there have been setbacks, it's also really unsurprising that car companies like Daimler, GM, and Ford are having trouble with self-driving vehicles. It would be like Vizio announcing that they were going to figure out quantum computing because they make displays. Building the car is the easy part, building the self driving part is the unsolved problem.

 

I honestly think that it's more of a question of risk than anything else. For an established company with a lot to lose, putting a self driving car out that crashes far less than a normal car is still a huge risk. I couldn't find any good recent numbers, but in 2016 Ubers were involved in at least 753 collisions that caused injuries and 10 fatal collisions in the US. If you have robots hurting hundreds of people per year and killing a few, even if you're a good deal safer than normal drivers, you still might not survive the public and regulatory scrutiny that would come with that.

 

So the barrier to entry isn't "is it as good at driving as a person," and probably something more like "is it 100x safer than a human driver."

[ThreePi]

Didn't that person who got hit and killed was because the car didn't expect a human outside a crosswalk? Like, that kind of goes to show how many lightyears they are from figuring this stuff out. What will they do in snowstorms when the road is obscured, or when snow or dirt is obstructing the sensors, or during construction when a route isn't clearly marked?

[AbsolutSurgen]

11 minutes ago, TwinIon said:

While I think it's clear that there have been setbacks, it's also really unsurprising that car companies like Daimler, GM, and Ford are having trouble with self-driving vehicles. It would be like Vizio announcing that they were going to figure out quantum computing because they make displays. Building the car is the easy part, building the self driving part is the unsolved problem.

Building cars isn't as easy as you think it is.  That's why most tech companies that hired automotive engineers, with a plan to engineer their own cars have backed away from it and are now trying to partner with car companies.

[TwinIon]

11 minutes ago, AbsolutSurgen said:

Building cars isn't as easy as you think it is.  That's why most tech companies that hired automotive engineers, with a plan to engineer their own cars have backed away from it and are now trying to partner with car companies.

I don't mean that building cars is easy, just that it's a solved problem that gives you no insight into building a self driving car. You can be the best car producer in the world, but that doesn't mean you don't have any leg up in building software and you don't have any of the manpower necessary to get working on it. They want to be working on it because it could fundamentally change their entire industry, but that doesn't make Ford/Daimler/etc aren't well suited for solving the problem.

[AbsolutSurgen]

2 hours ago, TwinIon said:

I don't mean that building cars is easy, just that it's a solved problem that gives you no insight into building a self driving car. You can be the best car producer in the world, but that doesn't mean you don't have any leg up in building software and you don't have any of the manpower necessary to get working on it. They want to be working on it because it could fundamentally change their entire industry, but that doesn't make Ford/Daimler/etc aren't well suited for solving the problem.

Putting an autonomous car on the road is a combination of hardware and software, and the entire system needs to be commercializabile (for lack of a better word):

1)  There are currently no Lidar systems that have low enough cost to be realistically implremetable

2)  The chips that are required to run an autonomous vehicle currently do not have the robustness to live up to a real life automotive environment

3)  The software isn't robust enough to sell to anyone (most developers are working on Level 4 autonomy, where the vehicles are limited to driving under significant constraints) -- No one has a near-term plan for Level-5 (which is what most people think about) where an autonomous car can drive anywhere

4)  The hardware costs (vehicle, computer system, autonomous hardware (i.e. the eyes of the system)) have insane costs

 

The car companies have several key core competencies:

1)  They have experience implementing computer systems in automotive environments (temperatures between -40 and +120 degrees F), salt, humidity, running for 12 hours a day, 365 days a year

2)  They have experience developing products with extremely low margins, working with hundreds of hardware suppliers to bring a product to market

3)  They already work with the fleet purchasers of these products, understand them, and have their trust

4) Have decades of experience of getting cars to actually run on the road

 

Software is ONE part of the solution.  Even if someone had a viable solution to the software issue.  They would still need to:

1)  Create a hardware computer platform that was implementable in a real world automotive situation

2)  Integrate that system with a real world automotive platform  (i.e. with all of the other systems that make a car run)

3)  Have a vehicle platform, that fleet customers can trust and want to invest in

 

Who is better suited to bringing a fully-functioning autonomous vehicle to market than someone like Volkswagen/Daimler/Ford/GM?  Google?  Only if they can find a partner that will do everything else, and leave them enough margin to make money.

[outsida]

The problem with self driving cars it that you have to spend 99% of your effort solving edge cases that represent 1% of the driving experience. The problem is not streamlinable, you need tons of real world data in order to arrive at an acceptable solution. That’s why I believe Tesla has the best chance of releasing a self driving car first because they have hundreds of thousands of people feeding them real data from autopilot. 

[Rodimus]

I just want to buy a self driving car so when I don't need it I can have it go out and uber for me. I would totally do that and not only have it pay for itself but bring in income.

[CitizenVectron]

16 hours ago, b_m_b_m_b_m said:

But I was told by well dressed people that we should plan the entire future of our cities, from transit to zoning to walkability, based on the idea that we will have driverless cars by 2020

 

No no, you're thinking of the segway.

[AbsolutSurgen]

12 hours ago, outsida said:

The problem with self driving cars it that you have to spend 99% of your effort solving edge cases that represent 1% of the driving experience. The problem is not streamlinable, you need tons of real world data in order to arrive at an acceptable solution. That’s why I believe Tesla has the best chance of releasing a self driving car first because they have hundreds of thousands of people feeding them real data from autopilot. 

They are the only major player trying to use cameras rather than Lidar.  The people I have talked to say this has no chance of ever working.  I have also seen autonomous vehicles do much more complicated things than what Tesla's "smart-summon" usually fails at.

 

The amount of data collected/used by a Lidar equipped autonomous vehicle is staggering.

[TwinIon]

15 hours ago, AbsolutSurgen said:

Putting an autonomous car on the road is a combination of hardware and software, and the entire system needs to be commercializabile (for lack of a better word):

1)  There are currently no Lidar systems that have low enough cost to be realistically implremetable

2)  The chips that are required to run an autonomous vehicle currently do not have the robustness to live up to a real life automotive environment

3)  The software isn't robust enough to sell to anyone (most developers are working on Level 4 autonomy, where the vehicles are limited to driving under significant constraints) -- No one has a near-term plan for Level-5 (which is what most people think about) where an autonomous car can drive anywhere

4)  The hardware costs (vehicle, computer system, autonomous hardware (i.e. the eyes of the system)) have insane costs

 

The car companies have several key core competencies:

1)  They have experience implementing computer systems in automotive environments (temperatures between -40 and +120 degrees F), salt, humidity, running for 12 hours a day, 365 days a year

2)  They have experience developing products with extremely low margins, working with hundreds of hardware suppliers to bring a product to market

3)  They already work with the fleet purchasers of these products, understand them, and have their trust

4) Have decades of experience of getting cars to actually run on the road

 

Software is ONE part of the solution.  Even if someone had a viable solution to the software issue.  They would still need to:

1)  Create a hardware computer platform that was implementable in a real world automotive situation

2)  Integrate that system with a real world automotive platform  (i.e. with all of the other systems that make a car run)

3)  Have a vehicle platform, that fleet customers can trust and want to invest in

 

Who is better suited to bringing a fully-functioning autonomous vehicle to market than someone like Volkswagen/Daimler/Ford/GM?  Google?  Only if they can find a partner that will do everything else, and leave them enough margin to make money.

The computer systems that car companies implement are largely built and maintained by third parties, especially if they have any actual computing power. Very little, if any of their computing expertise is applicable to what is required of self driving cars. Google does have a large chip making business, specifically focused on machine learning optimization. Ford's experience buying chips from BlackBerry RIM to run the UI for the explorer isn't really the same thing.

 

Margins don't matter when the product doesn't exist. Neither does the ability to work with fleet purchasers, or experience of getting cars to work on the road.

 

I will grant that when self driving cars do hit the road en-mass that they will likely be wearing the badge of an established car company. As you rightfully pointed out, building cars is hard, and it's not a business that Google or Uber or nVidia or anyone else is in a reasonable position to usurp. However, none of that experience really applies to the unsolved problems of self driving cars. The problems of vision and computation and machine learning optimization are not problems the car companies are positioned to solve.

 

Personally, I find it as hard to imagine that existing automakers "solve" self driving cars without licencing the tech as I do tech companies actually building their own cars.

[AbsolutSurgen]

3 hours ago, TwinIon said:

The computer systems that car companies implement are largely built and maintained by third parties, especially if they have any actual computing power. Very little, if any of their computing expertise is applicable to what is required of self driving cars. Google does have a large chip making business, specifically focused on machine learning optimization. Ford's experience buying chips from BlackBerry RIM to run the UI for the explorer isn't really the same thing.

Ford does not, and never has bought chips from RIM to run Explorer UI.

3 hours ago, TwinIon said:

Margins don't matter when the product doesn't exist. Neither does the ability to work with fleet purchasers, or experience of getting cars to work on the road.

 

Having a product on the road doesn't matter if there is no margin in it.

3 hours ago, TwinIon said:

I will grant that when self driving cars do hit the road en-mass that they will likely be wearing the badge of an established car company. As you rightfully pointed out, building cars is hard, and it's not a business that Google or Uber or nVidia or anyone else is in a reasonable position to usurp. However, none of that experience really applies to the unsolved problems of self driving cars. The problems of vision and computation and machine learning optimization are not problems the car companies are positioned to solve.

Argo AI (owned by Ford and Volkswagen) and GM Cruise are among the leaders in developing software for autonomous cars and have vehicles on the road.  They are currently behind Waymo.  Waymo, Argo AI and GM Cruise have all solved many of the problems associated with self driving cars -- only they know exactly where they are in the process, and how close they are to having something they can roll out in volume.

There are lots of chips that are available for sale to include in the logic unit for an autonomous vehicle -- I am sure that all of the automakers are working with a variety of different chip companies (Intel, nVidia, AMD, etc.) to provide those chips.

3 hours ago, TwinIon said:

Personally, I find it as hard to imagine that existing automakers "solve" self driving cars without licencing the tech as I do tech companies actually building their own cars.

I am sure there will be lots of cross-licensing of patents between the different autonomous vehicle manufacturers.

However, successfully commercializing an autonomous vehicle goes well beyond writing the software.  It is a necessary, but not sufficient condition.

These early autonomous vehicles will be very limited "Level 4" vehicles -- that will be geo fenced and have other significant limitations, and limited to commercial applications.  (Because they will be very expensive to produce, and run).