Autonomous Vehicles

[Glaucus]

Google Waymo announces it has racked up 4 million self driven miles

https://techcrunch.com/2017/11/27/waymo-racks-up-4-million-self-driven-miles/

Many believe the most important metric of building successful autonomous driving technology is actual miles driven on roads. Waymo continues to press its advantage.

Google began its self driven miles in 2009 and it took six years to accumulate its first 1 million miles. Waymo went from 3 million miles to 4 million miles in only six months (from May to November 2017). And the 4 million miles stat excludes the 2.5 billion that Waymo has "self driven" in its lab simulation.

Expect the launch of Waymo's commercial ride hailing service very soon.

tesla eclipses waymo. so it isn't an advantage.
also waymos test town with no safety driver is a bit of a cop out, as it has had every inch of the town mapped.
the other thing with waymo and many others is they are using lidar and although the price has come down it's still a large cost.

 

[sesevans]

Hi

A few observations:

1. Tesla is not offering Level 4 autonomy driving on its vehicles today, despite claiming its hardware and software is capable when turned on. Today their cars do offer Enhanced Autopilot and say: "Tesla’s Enhanced Autopilot software has begun rolling out and features will continue to be introduced as validation is completed, subject to regulatory approval." Therefore the Tesla driver always need to be vigilant when he or she operates in Enhanced Autopilot mode. The well-documented accident of a Tesla driver in the US last year, though not the fault of Tesla, provides a cautionary tale. Therefore, what Tesla offers today is not Level 4 autonomous driving.

2. Waymo has turned it on. The human "driver" is not asked to intervene. Waymo's 4 million miles in self drive mode has involved no driver intervention. Today they are providing Level 4 autonomous driving in a defined area.

3. Waymo states they place safety above all other factors. Unlike Tesla, the latest self driving vehicles "driven" by Waymo in Chandler, Arizona has no driver in the front seat. At the moment, the test driver sits in the rear of the car and not in the front seat. While the service is offered only to Waymo employees currently, it is expected soon to be offered as a ride hailing service in Chandler to the public. When Waymo is satisfied and has mapped the entire area of Phoenix (larger area than Greater London), they expect soon to expand their operating area. In time, there will be no driver in a Waymo vehicle.

4. It is not a "cop out" to carefully map in 3D a self driving area before offering a service to the public. If you place safety first, that must be done. Google is the pre-eminent provider of mapping, and the combo of Google Maps and Waze will prove very useful when the self driving ride hailing service expands.

5. I expect that ride hailing will be a very automotive intensive application. Currently the average vehicle is in use less than 5% of the time, sitting idle 95% of the time. A self driving vehicle with LIDAR will be used intensively, perhaps 50-60% of the time, with the balance used for battery recharging and any downtime. Therefore as a highly productive asset, the LIDAR equipped ride hailing vehicle will pay for itself quickly.

 

[Entai]

Only real issue at the moment is that Waymo is physically limited to 25mph, and rather to often for the companies liking drops into extremely cautious mode (10 mph) at complicated intersections, also it can only operate in an extremity limited fully mapped area.

Tesla had hoped by end of this year, but now say early next year they will demo a car driving from Los Angeles to New York without a single input from a human, including charging, and keeping to interstate speeds, to me that puts Tesla way ahead.

 

[Glaucus]

Hi

A few observations:

menas noothing, every single tesla rolling of the factory and have been rolling of teh factory for several years, drives in shadow mode whether its in auto pilot or not. it is gathering far more data than any other entity. This is extremely important.

you can not map the entire country and expect roads to never change, this is why waymo is cheating and why currently it can not roll it out.

three of your points has nothing to do with waymo and applies to all autonomous cars.

 

[sesevans]

and @Glaucus

Only real issue at the moment is that Waymo is physically limited to 25mph, and rather to often for the companies liking drops into extremely cautious mode (10 mph) at complicated intersections, also it can only operate in an extremity limited fully mapped area.

Tesla had hoped by end of this year, but now say early next year they will demo a car driving from Los Angeles to New York without a single input from a human, including charging, and keeping to interstate speeds, to me that puts Tesla way ahead.

Waymo is not limited to 25 mph. Some years ago, prior to Google spinning out Waymo, they launched the "firefly" self driving vehicle which had a top speed of 25 mph. There was a purpose to this vehicle. It was to garner confidence among residents, walkers, bicyclists and other drivers in the areas it operated. Today Waymo operates a fleet of more than 100 Fiat Chrysler Pacifica vans that travel at whatever the speed limit is in the four cities in which it operates: Austin, Texas; Spokane, Washington; Phoenix/Chandler, Nevada; and Mountain View, California. Prior to the firefly and for many years, Google operated its self driving vehicles on motorways but determined that it was more challenging to operate their self driving vehicles in suburban traffic in which bicycles and pedestrians were prevalent as well as other vehicles.

I appreciate what Tesla has said about its plans to operate a self driving vehicle across many State lines but there are still many US States that do not permit such a vehicle. Currently there is a bill in Congress to pass regulation of self driving vehicles mostly to the Federal Government, leaving the States with minimal regulatory powers. Operating on a motorway or Interstate is much less impressive than it sounds and it is why Google moved its fleet away from such operations years ago. What does impress me however is Tesla's ability to operate in such a potential scenario on one battery charge! If what you say is correct about the battery, which currently has a much shorter range, that would be a big breakthrough in EV battery technology and less impressive in the area of autonomous driving.

No other manufacturer or operator has accumulated 4 million miles of self driving vehicles without human intervention apart from Waymo. Furthermore, they have 2.5 billion miles of lab tested mileage which informs their road operated self driving vehicle too.

As to gathering data, it remains to be seen how Tesla actually uses the data it has gathered. Just because they have gathered data does not mean that they have applied it to a Level 4 autonomously driven car, which is something Waymo has actually done.

There has never been a fleet of fully self driving cars on public roads until now with Waymo---no driver behind the wheel----and @Glaucus, no cheating. Watch the video:

https://www.youtube.com/watch?v=aaOB-ErYq6Y

 

[Glaucus]

yes cheating, waymo highly map the area before setting lose autonomous cars, something they can not do for a general roll out.
tesla has gathered far more miles of autonomous miles and billions more of general driving miles, which goes into their deep learning system.

waymo 4million, tesla a year ago already had over 100million miles of autonomous driving and billions of miles n shadow mode.

 

[sesevans]

yes cheating, waymo highly map the area before setting lose autonomous cars, something they can not do for a general roll out.
tesla has gathered far more miles of autonomous miles and billions more of general driving miles, which goes into their deep learning system.

waymo 4million, tesla a year ago already had over 100million miles of autonomous driving and billions of miles n shadow mode.

This recent post in Wired sums up the pros and cons about Tesla's approach. By the way, Waymo's deep learning system is formidable too.

https://www.wired.com/2016/10/teslas-self-driving-car-plan-seems-insane-just-might-work/

 

[Glaucus]

This recent post in Wired sums up the pros and cons about Tesla's approach. By the way, Waymo's deep learning system is formidable too.

https://www.wired.com/2016/10/teslas-self-driving-car-plan-seems-insane-just-might-work/

didnt say it wasnt formidable, but unlike what you suggest it has a fraction of the data to work with than tesla, then two big hurdles, a) it relies on heavily mapping the town which defeats the point and worsens the ai. and second it relies on expensive lidar.

tesla has a cheaper system with far more data, and despite a lot of outlets and people being negative about Tesla for financial reasons, experts in the field are starting to come around and say vision-based systems are the future and all agree that data is king and tesla has much more data than anyone else out there and not by a small amount.

Does that mean others cant make self-driving cars, absolutely not, but i wouldn't put them as the most likely to be the first to mass produce self-driving cars.
Tesla produced 100,000 thousand cars this year with the equipment on all driving and collecting data, next year tesla will likely produce around 500,000 cars, small test sites like waymo cannot produce even remotely that much data.

 

[sesevans]

didnt say it wasnt formidable, but unlike what you suggest it has a fraction of the data to work with than tesla, then two big hurdles, a) it relies on heavily mapping the town which defeats the point and worsens the ai. and second it relies on expensive lidar.

tesla has a cheaper system with far more data, and despite a lot of outlets and people being negative about Tesla for financial reasons, experts in the field are starting to come around and say vision-based systems are the future and all agree that data is king and tesla has much more data than anyone else out there and not by a small amount.

Does that mean others cant make self-driving cars, absolutely not, but i wouldn't put them as the most likely to be the first to mass produce self-driving cars.
Tesla produced 100,000 thousand cars this year with the equipment on all driving and collecting data, next year tesla will likely produce around 500,000 cars, small test sites like waymo cannot produce even remotely that much data.

We clearly have a difference of opinion about the value/quality and quantity of useful data that Waymo has collected over 9 years of efforts in autonomous driving. Also having professional testers on board their autonomous vehicles until now has increased the quality of the data collected exponentially. There are many joint ventures and partnerships that Waymo can form that will fill in any missing part of their asset-light approach (as opposed to Telsa's asset-heavy approach). Already they have partnerships with Fiat Chrysler, Avis, Lyft and others.

As a future rider in an autonomous vehicle, I would always value safety and the value of having LIDAR in addition to all the other sensors and radar on Waymo vehicles makes me feel more secure. The expense issue you raise is a red-herring. If the ride hailing vehicle is operating 50-60% of its useful life, LIDAR will be seen as a low cost addition. Google has already said that they can produce LIDAR for less than $ 10,000 and expect it to drop significantly from there as it is mass produced. For the moment, let's politely agree to disagree.

 

[Amp34]

Only real issue at the moment is that Waymo is physically limited to 25mph, and rather to often for the companies liking drops into extremely cautious mode (10 mph) at complicated intersections, also it can only operate in an extremity limited fully mapped area.

Tesla had hoped by end of this year, but now say early next year they will demo a car driving from Los Angeles to New York without a single input from a human, including charging, and keeping to interstate speeds, to me that puts Tesla way ahead.

Driving on open roads (which is what that trip will largely be) is a lot easier than in towns. Outside of a few areas level 4/5 is not going to be coming to the general public in the next decade. It’s going to come a lot faster on the highways and in mapped towns (for taxis) but elsewhere it’s still a long way off. It doesn’t matter if it’s Tesla or Waymo (or any others).

There are three options with LIDAR at the moment. Tesla’s way, which involves just not using it, there’s the Cadillac (and others) way, which is mapping major roads and providing that data to the cars computer, but not actually having lidar on board, and then the third way which is mapping the roads AND having Lidar on vehicles.

Realistically for level four automation in the near future on more complex routes you need that accuracy, whether it’s preinstalled or on the fly. In the future it probably won’t be as much of an issue. It’s certainly not “cheating”.

 

[Entai]

and @Glaucus

What does impress me however is Tesla's ability to operate in such a potential scenario on one battery charge! If what you say is correct about the battery, which currently has a much shorter range, that would be a big breakthrough in EV battery technology and less impressive in the area of autonomous driving.

Who said anything about them doing it on one charge ??

They said that charging will not need human input, that is what they are working on and will change the face of EV completely.

 

[sesevans]

Who said anything about them doing it on one charge ??

They said that charging will not need human input, that is what they are working on and will change the face of EV completely.

OK so you are saying the breakthrough in EV technology is that a robot charging your battery rather than you getting out of the car to initiate and terminate the (20-30 minute?) process will "change the face of EV completely."

Really? I think you have a low bar for what is a mega-breakthrough.

I would have agreed with you (had it been true) that the 3,000 mile no-need-for-a-charge would have been a mega breakthrough.

 

[sesevans]

Anyone following the Waymo trial discovery against Uber over Uber's alleged theft of Waymo self driving car trade secrets?

Uber had a horrible day in court yesterday. Uber's business practices are at best highly questionable and at worst illegal.

https://www.wired.com/story/hidden-evidence-uber-waymo-self-driving-car/

 

[sesevans]

I can just see the dangers a self driving thing brings. I'd rather have a person in charge and control that brings.

You mean the "person(s) in charge and the control that brings" to the 1.3 million road traffic accidents that occur globally every year, according to the World Health Organisation?

Think about that number for a moment and ask yourself: if a solution can be found which drastically reduces this annual loss of life, shouldn't the world be extremely interested in developing it and bringing it into production?

 

[sesevans]

General Motors held an investors conference today in California to highlight their progress with autonomous vehicles. They own a start-up in California called Cruise Automation where the branding for their AV comes from. They expect to launch their ride sharing service in US urban cities in 2019.

Their slide show from today's conference is attached and will assist us in our discussion of this fast changing field.

GM made the following statements about their AV plans:
1. Self driving Chevy Bolts could be manufactured at scale at GM's existing plants, driving down costs and rapidly deployed in major metropolitan markets through a ride service to disrupt incumbents (Uber and Lyft). I note that GM is also a shareholder in Lyft so it is potentially competing with its own investment.
2. GM sees deployment of AVs in quarters, not years.
3. Biggest opportunities initially will be on US East and West Coasts.
4. They believe a robo-taxi service could be potentially bigger than their current core business.
5. They seek to remove 40% of the cost of current ride services, expecting to lower the cost to below $ 1 per mile by 2025 from about $ 2.50 per mile today, with margins of 20-30%.
6. They expect a robo-taxi service could increase their sales significantly as the vehicle would be continuously used by multiple users over its lifespan.
7. As an example of bringing down the cost of AVs, GM said it expects to lower the cost of long-range LIDAR sensors to $ 300 from $ 20,000. Last month GM bought LIDAR specialist, Strobe Inc.
8. GM expects to sell 1 million EVs a year by 2026.

https://www.gm.com/content/dam/gm/events/docs/5265893-685163-Chartset-11-30-2017

Comments?

 

[Glaucus]

and tesla is in talks with Nevada and California to run driverless lorries between their two plants.
Despite what some think, driverless is coming.

 

[BananaDunka]

I really like the idea. Whilst ICE cars will carry a premium in keeping them on the road in the future, at least those who are petrolheads will have an easier life driving on the roads in a time when autonomy is the norm. Those who don't have a crap about driving standards these days will more than likely switch to autonomous vehicles taking the risk of an incident due to human error away. The obvious downside is insurance and everything else may go up for those who wish to remain "analogue".

Swings and roundabouts I guess.

I'd definitely buy an EV if it looked like that Honda concept. Looks cooler than any other EV I've seen in recent years.

 

[PMKeates]

Those who don't have a crap about driving standards these days will more than likely switch to autonomous vehicles taking the risk of an incident due to human error away. The obvious downside is insurance and everything else may go up for those who wish to remain "analogue".

Swings and roundabouts I guess.

Not quite. If autonomous cars are very safe and the majority of unsafe drivers switch to them, then what’s left are the safe drivers. The net result will be fewer accidents involving the remaining drivers and near zero accidents for the previously unsafe ones. This will mean lower premiums for everyone involved, including the remaining drivers.

 

[BananaDunka]

Hmm that's fair, but I don't have much faith in a system that just seems to always been on an upwards source of increasing costs for everything.

 

[sesevans]

While it is too early to state definitively that auto insurance costs will be coming down, an interesting note from Munich Re written last year discusses the implications for the insurance industry and the potential for pricing to drop for customers. A few interesting points to highlight from this report include the likelihood that most of the insurance cost of a Level 4 AV will pass to the manufacturer of the vehicle and third party suppliers, esp where in most use cases of such AV's we are looking at ride sharing vehicles owned by large fleets of companies/fleet operators and not individuals. I suggest that these large fleet operators will have significantly greater bargaining power with insurance underwriters than you and me on an individual basis.

The key for the underwriter in assessing risk and pricing will be the ability to explore and use better vehicle data being generated. Current laws make this difficult. As the report points out, "Regulators currently prohibit insurers from using certain data in their rating models, including location, speed, or other data considered private. This limits the usefulness of the data captured by the AV. The more insurance companies must rely on traditional pricing information, the more insurance is likely to remain the same."

I can imagine that the willingness to change regulations will vary with geographical location. In China, if the State mandates the change, insurance pricing will benefit by the ability to use increased data. Privacy means much less in China than in the West. I see the US being next likely to change to accomodate fleet owners sharing of the data with insurance companies. In the EU where privacy seems to be "king", I see changes coming more slowly. It will be interesting to see what Britain does on this issue as it negotiates its exit from the EU, esp with its interest in expanding AV use on British roads.

The data that a Level 4 AV provides will be huge and the owner of the data should, in my opinion, be free to exploit its commercial use. In the GM conference presentation on AV's this week in California, GM company management mentioned that they hope to be able to use captured data (video, HD maps, travel patterns, etc) and make it available to third parties (eg insurance companies). For example, in the US, the motor insurance losses annually are $ 1 trillion of which $ 100 billion is the estimated amount of fraudulent claims. If captured data can be sold to insurance companies, fraud rates can be dramatically lowered with the potential for a big drop in insurance losses that can affect everyone's insurance pricing. In another example cited, lane level traffic info can be commercially beneficial, even though this is not an insurance company issue so much. Traffic congestion in the US is estimated to cost operators approximately $ 100 billion per year in lost productivity and time. Real time info here could yield large benefits to operators who would be prepared to pay the data owner for access to this info.

https://www.munichre.com/site/mram-...a/PDFs/3_Publications/Autonomous_Vehicles.pdf