Tesla Autopilot Competition Developments

[replicant]

I'd like to start a thread related to the competition to Tesla' AP, now that Tesla aims at full autonomy within 2 years.

Note that a thread already exists for competition in battery-electric vehicles.

- - - Updated - - -

BBC: Ford says it has conducted successful tests of its driverless cars in snowy conditions.

Ford said it had [...] programmed the Lidar sensors to detect landmarks above the ground, such as buildings and road signs. The car could then compare this information to an existing high-resolution map of the road - generated by autonomous cars during more favourable weather- stored in its computer. This allowed the car to navigate the road even when visibility was poor.

 

[replicant]

Scenario of the Tesla competition with fleets of autonomous cars.

Let say the Model 3 will be sold with Autopilot and free Supercharging as options.
By 2020, the Model 3 with AP and the revamped Models S and X will fully autonomous.

Tesla will announce its on-demand transportation service only when:
- its cars drive 100% autonomously, anywhere and without human intervention.
- annual sales reach some big numbers (500K?) so that Tesla has enough cash to build new gigafactories and assembly lines

With Tesla on-demand service, customers will have to option to drive the cars themselves (for a fee). This will make Tesla more desirable than the Google self-driving pods. Tesla will keep selling cars to individuals and transportation companies (like Uber) in order to get as much cash as quickly as possible.

When Tesla generate enough cash to build dozens of factories without significantly diluting its shareholders, most of the cars produced will be added to the Tesla fleet.
As customers understand the benefit of on-demand transportation (cheaper and available instantly anywhere), Tesla will offer its car owners the option to repurchase their Tesla, and add them to the Tesla fleet.

Then, Tesla will announce the 4th generation platform to best compete with other self-driving fleets (human won't be able to drive the cars from this platform).

Around the same time, Apple will launch a fully autonomous cars.
Like Tesla, the car will be designed by Apple and under its brand. But it will be manufactured by OEM partners (like with Foxconn for the iPhones). BMW as first possible partner.
Contrary to the Model 3, the Apple Car will only be available for on-demand transportation (won't be sold to consumers).
Individuals and companies will be able to use the car as a on-demand delivery platform.
Shops, malls and employers will be able to participate in the Apple Charging program (similar to Tesla Destination charging).
With the company advertising and investment in factories (for car and batteries), Apple will quickly become the most well-known on-demand transportation platform in the biggest cities around the world.

Google will launches its self-driving car fleet a little sooner than Apple. Google will have its own fleet of cards (like Tesla and Apple) but it will allow auto partners to manufacture their own autonomous. These cars will use Google's self-driving software freely. Google will allow partners to monetize their fleets of cars: they'll just have to join the Google On Demand Transport Network (exclusively). The project will be similar to Android: with AOSP partners for OEMs, Google Play Services for the Google's Transport Network, the Nexus program for cars designed and manage 100% by Google.

Amazon will be late to the party and try to combine its program with its drone delivery project. It will enter the market with other OEMs to directly compete with Apple and Google.

Uber will also have to partner with auto-manufacturers to switch to autonomous vehicles (human drivers will be too expensive pretty quickly).
The company will probably prefer Tesla to traditional manufacturer, as the giants (Google and Apple) will become huge threats with their existing user base, their industrial expertise and their cash reserve. The two companies might consider a merger to better compete with the Silicon Valley giants.

Microsoft might stay on the side and let its competitors fight for this market while it focuses on the information industry. Alphabet might be distracted and Microsoft and Facebook hopes to take advantage of the situation.

More importantly, the traditional automakers will be forced to:
- switch to 100% battery-electric vehicles in just a few years
- implement OTA and autonomous driving
- abandon the dealership model at a cost
- consolidate 'in a rush': as people abandon their cars for on-demand transportation, car production will temporarily increase to make big fleets of self-driving cars for Google and Apple, but then the market industry will concentrate around a few leaders.
- invest in charging network to avoid become more dependent on actors like Tesla and Apple.

At some point, independent charging networks will emerged with a standard to charge anywhere (cf. USB), for a fee (cf. payment card systems).

What do you think?

 

[replicant]

Bill Ford Isn’t Scared of Apple: Henry’s great-grandson explains how the automaker can become a software-driven service company that makes cars, too

 

[dakh]

Few things that I question here.
1. You don't need full autonomy everywhere to start making bank. Just West Coast city limits (no snow, no high-speed driving) would be huge already. Just highway driving to next exit is pretty substantial too.
2. It looks much more logical to me that the "rental fleet" would be a somewhat different car from any existing models. There's a big difference in usage pattern. It won't be anything drastically different, but some changes geared towards commercial use.
3. I see a big advantage in Tesla's fleet of cars out there that are actually driven by real people, here's why. Whoever would be trying to get their solution approved for general use by regulators would have to have some strong evidence that the system work, and is safer than human drivers on average. All these Google/Apple lab cars driving around, I think would be harder to make a case for regulators based on those vs. Tesla's actual customer vehicles stats. So I think to answer the question of when all these players would get the actual product out that can be used would depend a lot on regulatory environment. Could be that say Tesla is first to make this work but regulations are slow to change so others catch up while Tesla is waiting for permission to operate their solution. Or the other way around. Or say California or even just Bay Area would issue permissions first but require certification process. Hard to tell the timeline here.
4. Shared fleet is just one facet of providing the whole "global mobility" solution, and I think focusing too much on just this is also short-sighted. What about say a giant rock concert on the country side? You gotta have a fleet of high capacity vehicles that are smart enough to predict demand, aggregate people at transfer spots, know to keep those vehicles at the venue until the show is over, etc. So on-demand city transport is like the first computer program, very crude but brings great benefits. There's a lot more to be had past that though.

 

[schonelucht]

Google's autopilot isn't perfect (yet). According to the company they had 13 situations when a human operator intervened without which an accident would have been guaranteed. Operators intervened more often but simulations showed that there likely wouldn't have been an accident had the car continued on autopilot.

 

[electracity]

Are any of the autopilot car control functions functions actually done with Tesla developed software? Tesla is probably feeding vehicle position estimates, in conjunction with Garmin software, to Mobile eye software. But it is unlikely that somehow tesla developed software is sharing car control with mobile eye software.

Tesla undoubtedly has software developers, and has said they are increasing their in house effort. But there is substantial evidence that it is Mobile Eye driving the car.

[FONT=Helvetica, Roboto, Segoe UI, Calibri, sans-serif]"Ramping up the Autopilot software team at Tesla to achieve generalized full autonomy. If interested, contact [email protected]."[/FONT]


[FONT=Helvetica, Roboto, Segoe UI, Calibri, sans-serif]It is unlikely that Mobile eye would enter into an exclusive agreement with Tesla on any major portion of their technology. So what is the path to Tesla becoming preeminent in autonomous driving?[/FONT]

 

[dakh]

We've had discussions on this before. I got an idea of what's going on but haven't seen anything compelling to prove or disprove it. In short, MBLY is responsible for interpreting images and creating an on-going 3d model of what is out there (including recognizing pedestrians, road signs etc.). TSLA is responsible for everything else, the bulk of which is building a dynamic model of what is happening and making control decisions based on that. Building 3d dynamic model might be in MBLY's court, either way would be a valid solution. Could be that's the area where they cooperate. Bottom line is I suspect control is in Tesla's realm, and modeling the world in MBLY's. And also looks to me MBLY is working on the control part on their own. Plus there's overlap in, uh, overlapping map data onto the 3d model. Telsa got data from its cars, MBLY has built or is building a similar system.

 

[dakh]

To clarify, when i say dynamic model i mean predictive one.

 

[replicant]

To clarify, when i say dynamic model i mean predictive one.

I highly recommend this 2016 presentation by MBLY's CTO:

It covers everything, from MLY's technology, strategy, partnerships and competition.

3 pillars to reach full autonomy:
- sensing: produce 3D environmental model with camera, radars and lidars to indentify space and objects (moving and stationary)
- driving control: predict best path by doing supervised learning with portable Deep Neural Networks. It's very difficult to predict the future in a multi-agent environment.
- mapping: localization and very high resolution maps

2 mapping strategies:
- Google's way: low-res sensing + high-res 3D maps (available in limited areas at first)
- Mobileye's way: high-res sensing + low-res 3D maps (works everywhere from the start)

What's Mobileye is doing:
- 360° sensing: use 5 cameras + 1 trifocal camera + radar/lidar for redundancy
- mapping: localize the vehicle with 10 cm longitudinal and 5 cm lateral accuracy in a road model. It crowd-sources a sparse 3D model (with major landmarks only) and create a dense 1D model (by recording the polynomials of the lanes). The "roadbook" is created in realtime and requires a very-low bandwith compared to Google high-res maps (10 kb/km).

 

[ecarfan]

Interesting. I wonder if Google would agree that the Lidar and camera system they use is "low-res sensing". I think not.
It seems that both mapping strategies can work. Of course right now you can buy cars with limited self-driving capabilities based on Mobileye's approach, and Google has yet to sell any cars. But Google does not do a good job of commercializing their hardware projects. They tried selling a phone, and then stopped. Their Chromebook has seen limited sales compared to the competition. Glass is not yet for sale.

 

[blablub]

How does Teslas crowd-cources mapping-data differ from mobileyes "roadbook"? Could it be the same, because tesla uses mobileyes EQ3 chip.

 

[electracity]

I highly recommend this 2016 presentation by MBLY's CTO:

I hear a guy who has no idea how he will get to full autonomy without strong A.I. Not one company (AFAIK) has chosen cameras who are doing pure autonomous driving development research. He never points out that his high resolution cameras are passive devices, and that "low resolution" LIDAR are active.

The last four minutes are interesting. I think he is clearly saying that Tesla and Mobileye have become frenemies.

I doubt very much that either Google or Apple will have any problem doing exact car position with low bandwidth. Google probably is currently doing high bandwidth networking to google car. This situation says nothing about their deployment plans. Google does know a few things about efficient mapping and networks.

- - - Updated - - -

How does Teslas crowd-cources mapping-data differ from mobileyes "roadbook"? Could it be the same, because tesla uses mobileyes EQ3 chip.

Probably accesses the same raw data stream, but probably vary in how they handle the data. I presume now Tesla now feeds position data back to mobileye in 7.1 firmware. Mobileye is pursuing a parallel/competitive development path with GM and others. Tesla's development will presumably try to remove mobileye from future vehicles.

 

[RobStark]

GM delays Super Cruise rollout

In a statement to Automotive News, GM confirmed that the so-called Super Cruise technology will debut on the CT6 large sedan sometime during 2017. That would be at least several months later than the fall 2016 on-sale date that GM CEO Mary Barra cited when she announced the plans in September 2014.

http://www.autonews.com/article/20160112/OEM04/160119857/gm-delays-super-cruise-rollout

 

[replicant]

Google Self-Driving Car Project

For the past few years, members of our team have been working from the Greater Detroit area. Now it’s time to lay down roots: we’re establishing a self-driving technology development center in Novi, MI.

Many of our current partners are based here, so having a local facility will help us collaborate more easily and access Michigan's top talent in vehicle development and engineering. At this 53,000 sq ft development center, our engineers, working with local partners, will further develop and refine self-driving technology. One of the first tasks will be to ready our self-driving Chrysler Pacifica hybrid minivans.

We’ll begin moving in throughout 2016, but as you can see, there’s some work to do before it feels like home!

 

[electracity]

Google Self-Driving Car Project

Google may be in Detroit now, but if that company creates a superior solution, I bet that it appears first in a Tesla. Look at GM trying to roll out mobileye in supercruise. Sleepytime.

 

[dakh]

From the Code interview when grilled on further Model 3 info and autopilot, Musk said "We're gonna do the obvious thing". Do I read this right which is it'll be all sensors included, safety features enabled but gotta pay $$$ for autopilot feature?

 

[Saghost]

From the Code interview when grilled on further Model 3 info and autopilot, Musk said "We're gonna do the obvious thing". Do I read this right which is it'll be all sensors included, safety features enabled but gotta pay $$$ for autopilot feature?

Seems likely to me. The other interesting tidbit, from the 2016 shareholder meeting the other day, was Elon's comment about the S and the X always being the place for new technology.

Sorta sounds to me like the 3 will initially come with current generation Autopilot or something slightly improved, and the S and X of that time frame will have the full autonomous package hardware, if not the software yet.

 

[dakh]

3 vs S probably will depend on the price of AP hardware, if throwing more money at it helps achieve full AP then yes S will have it before 3. It could also be that by the time 3 ships they'll have confidence in the AP ability of an inexpensive suit so it'll just ship with the same hardware.

 

[CButterK]

May’s self-driving car report details how Google teaches its cars to honk

So... Google car can honk all by itself now. What can Tesla do today?

Why do I get this feeling that Tesla's Lead is getting at least a Million Miles per day farther than the competition?

 

[lklundin]

True to its aspiration of having a hand in anything IT-related, Microsoft has now expressed its intent to also "enable autonomous vehicles and assisted driving":

Microsoft wants to power self-driving cars, not build one

Apart from the obvious safety concerns (e.g. BSOD and malware) that would turn into actual life-or-death scenarios, I cannot imagine an automaker who would pay (or pass on to their customer) license fees for an operating system when Linux is readily available and already in automotive use (e.g. by Tesla Motors).

And I have to wonder: What automaker would accept a sticker with "Microsoft inside" on the cars?