Autonomous Vehicles

[sesevans]

Sounds like a subtle way of saying it's not as safe as they thought it was

While they can become safer, it would be unwise to not appreciate the strides they have made in a relatively short period of time.

This week, Waymo executives cited a project from several years earlier, where researchers used Google’s powerful neural networks to reduce pedestrian detection errors from about 1 in 4 to about 1 in 400, in a matter of months. The company didn’t say what its error rate was currently, but it would obviously have to be significantly better than 1 in 400 by now to be ready for public roads (it is the only company in the world operating its fleet of AVs without a safety driver and with passengers).

They went on to say: “Cars have lasers to measure distance and shape, and radars to measure their speed. By applying machine learning to this combo of sensor data we can detect pedestrians in all forms in real time.” Waymo’s self-driving technology has been refined to it can detect pedestrians even if they are hauling plywood or wearing dinosaur costumes.

Progress.

 

[sesevans]

Toyota sees their rivals Google and Apple as existential threats to their franchise and responds by pouring in $ 22 billion into R & D in this fiscal year.

https://asia.nikkei.com/Business/Companies/Toyota-pours-22bn-into-R-D-as-Apple-and-Google-close-in

 

[sesevans]

Will self driving vehicles need 3D maps of the entire country or is there an alternative using GPS?

https://www.geospatialworld.net/blogs/mit-maplite-self-driving-cars-hd-maps/

 

[sesevans]

Financial Times today: "Waymo forecast to capture 60% of driverless market"

The FT has an article today, based on research from investment bank UBS: "Waymo forecast to capture 60% of driverless market". UBS says " Waymo will own 60% of the global self driving taxi market by 2030, a level of dominance that forces many of the world's carmakers to adopt its technology or become obsolete." UBS estimates global revenues from self driving tech by 2030 will be up to $ 2.8 trillion, with Alphabet's Waymo unit the global leader. They say "only a select handful of carmakers such as Daimler and GM will be able to operate their own systems and compete with Waymo's tech, following interviews with self driving developers, technology groups and academics as well as UBS' own analysis."

UBS expects that 12% of cars sold in 2030 will be for driverless taxi fleets, with a total of 26 million robotaxis in operation. Private car sales will fall by 5% as a result. The report predicts that demand for self driving taxis will take off around 2026, depending on public acceptance of the technology in the face of recent crashes, and regulatory approval, although it will develop at different speeds in different markets. UBS believes the largest revenue pools will be in operating the car booking networks and monetising time spent by passengers in the cars, with building the cars and other services such as mapping or sensors taking a smaller portion of the driverless pie. But the costs of building a self driving system from scratch, as well as the challenges of deploying it in cities across the world, will prohibit all but a handful of carmakers from competing in the most lucrative part of the market, according to UBS.

UBS says that "unlike most auto players, Google focused on fully self driving technology from the very beginning, more than 5 years before the auto industry started working on it. Waymo has notched up more than 6 million miles of physical testing in California and elsewhere in the US as well as 5 billion miles of virtual testing on its computers, putting it far ahead of rivals. Analyst Patrick Hummel, who led the UBS report, said: "Very few players other than Waymo will be able to succeed with having an autonomous vehicle brain in the marketplace." "Sooner or later carmakers that are unable to compete will have to give in and take the Waymo system ." And "those that do provide vehicles for driverless fleets may see their brands disappear from the cars, and be relegated to "white label" providers."

Over time he concludes this will lead to a shake out in the car industry and a reduction in the number of carmakers. There are dozens of carmakers around the world in a fairly fragmented industry. The majority will be in the loser category.

Prediction: in a couple of years, the Overclockers Motor threads will be predominated by autonomous vehicles!

Any thoughts?

 

[jpaul]

This week, Waymo executives cited a project from several years earlier, where researchers used Google’s powerful neural networks to reduce pedestrian detection errors from about 1 in 4 to about 1 in 400, in a matter of months. The company didn’t say what its error rate was currently, but it would obviously have to be significantly better than 1 in 400 by now to be ready for public roads (it is the only company in the world operating its fleet of AVs without a safety driver and with passengers).

not sure 1 in 400 is anything to write home about .... all part of the clever marketting ploy to confer an aora of safety ?

Reading through this paper
https://arxiv.org/pdf/1706.04599.pdf
which shows neural net error rates on detection of a number of interesting image datasets

1. Caltech-UCSD Birds (Welinder et al., 2010): 200 bird species. 5994/2897/2897 images for train/validation/test sets. 2. Stanford Cars (Krause et al., 2013): 196 classes of cars by make, model, and year. 8041/4020/4020 images for train/validation/test. 3. ImageNet 2012 (Deng et al., 2009): Natural scene images from 1000 classes. 1.3 million/25,000/25,000 images for train/validation/test. 4. CIFAR-10/CIFAR-100 (Krizhevsky & Hinton, 2009): Color images (32 × 32) from 10/100 classes. 45,000/5,000/10,000 images for train/validation/test. 5. Street View House Numbers (SVHN) (Netzer et al., 2011): 32 × 32 colored images of cropped out house numbers from Google Street View. 598,388/6,000/26,032 images for train/validation/test

https://www.cs.toronto.edu/~kriz/cifar.html
Superclass Classes
aquatic mammals beaver, dolphin, otter, seal, whale
fish aquarium fish, flatfish, ray, shark, trout
flowers orchids, poppies, roses, sunflowers, tulips
food containers bottles, bowls, cans, cups, plates
.....

some of results are in the 0.25% domain, 1 in 400 ... (thats distinguishing/classifying a tulip from an otter etc etc)

.. executive BS?


edit : the car should be identifying the age, ethnic origin, gender of the pedestrians

 

[sesevans]

not sure 1 in 400 is anything to write home about .... all part of the clever marketting ploy to confer an aora of safety ?

...........

I think the operative words in the paragraph you cite are: ".....significantly better than 1 in 400 by now.....

Why? Because they plan to bring a commercial ride hailing service to Phoenix, Arizona "later this year."

 

[sesevans]

At a recent NY conference, GM sees custom designs, personal ownership of self driving cars by 2023-2025.

http://www.autonews.com/article/20180510/MOBILITY/180519970/gm-autonomous-vehicles-customization

 

[sesevans]

Nvidia CEO says pause in public road testing will end soon now that Uber has released the reasons for the fatal Uber crash (which uses Nvidia solution).

https://www.marketwatch.com/story/n...ssues-did-not-affect-auto-business-2018-05-10

 

[jpaul]

now that Uber has released the reasons for the fatal Uber crash

sorry did I blink ?
one source said something about a plastic bag causing confusion ?

The Information reported on Monday that Uber has determined the likely cause of the collision in March that killed a pedestrian was a problem with the software that decides how a self-driving car should react to objects it detects. The outlet said the car’s sensors detected a pedestrian but the software decided the car did not need to react right away.

“We can’t comment on the specifics of the incident,” Uber said regarding the report, citing an ongoing investigation by the National Transportation Safety Board (NTSB)

so nvidia are being a bit disingenuous; but it's in their interest to push it under the carpet.

 

[sesevans]

sorry did I blink ?
one source said something about a plastic bag causing confusion ?


so nvidia are being a bit disingenuous; but it's in their interest to push it under the carpet.

Interesting comment.

I was listening to Nvidia conference call. They did not discuss restarting of testing in that call. They made other observations about what they believed to be the size (TAM) of their GPU solution for the automobile market. They believe in time every vehicle that moves will be autonomous: cars, robotaxis, tractors, trucks, etc. They see the GPU TAM at $ 60 billion by 2035 with more than 100 million personally owned AVs and more than 10 million robotaxis on the road. For example, they said that the GPU inference need in a Level 2 vehicle might be $ 200 in value to them while the GPU inference value in a Level 5 vehicle would be more like $ 2000. They believe their GPU solution is ahead of Google's TPU 3 announced by Google this week.

Internet of Business.com blog has a note referencing the UBS research I discussed in an earlier thread (#985) and agrees with their conclusion about Waymo's eventual market share but only among Western Nations as it sees China (particularly Baidu) as having natural advantages in the China domestic market. You might find their blog of interest. Do you agree with their thinking? I appreciate that China tightly controls (some say illegally so) their domestic market against outside competition and the current trade dispute on intellectual property, etc may open things up for companies like Google/Waymo in due course but maybe not.

https://internetofbusiness.com/waymo-driverless-car-leader-2030/

 

[Malevolence]

Just plonk a Recaro on top.

 

[Nasher]

Maybe Nvidia's chip was missing a portion of it's memory and fell over

 

[jpaul]

sorry did I blink ?
one source said something about a plastic bag causing confusion ?

I was being facetious - that one unacredited source floats a theory about the bag, which uber does not comment on, and nvidia, then takes as gospel
(the source wants a 20 minute email address before I could read the article too)

 

[sesevans]

Waymo seen bolstering its technical expertise with US Federal authorities with hiring of senior Tesla executive.

https://arstechnica.com/cars/2018/0...al-contact-with-the-feds-just-left-for-waymo/

One of the comments in the Ars link I found interesting:

"From my own experience as a one time company contact with regulatory bodies and standards makers (in the field of electrical safety), I wonder if this has a very simple explanation. Waymo is committed to LIDAR. I suspect that technical bodies and regulators are heading down the route that all self driving cars must have it. Tesla here has a huge problem; the cars they sold that were claimed to have a software-only self driving update capability. If LIDAR becomes mandatory for self driving cars, they are stuffed. A retrofit is more or less unimaginable and would certainly cost much more than $4000. In order to contest such a regulation Tesla would have to show it had no technical merit.
If I'm right about this Schwall would be crazy not to jump ship if he had the chance. If he stays and LIDAR is mandated, Musk will be throwing things at him for failing to prevent it. If he goes to Waymo he can use his contacts - important in the standards and regulatory world - to get the desired outcome which will benefit his new employer and seriously embarrass his old one. And he's the virtuous guy who went to the company that could deliver self driving cars."

 

[jpaul]

this article is getting some attention in the uk

AI expert warns of self-driving car dangers - 'keeping them under control while in the wild not possible'
Senior British scientist who is one of the leading minds in artificial intelligence (AI) has warned against self-driving cars, arguing they are not safe because engineers cannot predict how their creations will behave “while in the wild”.

Demis Hassabis, co-founder of the AI specialist DeepMind, acquired by Google for £400m four years ago, urged caution despite his own role in cutting edge research on thinking machines.

Speaking at the Royal Society in London, Dr Hassabis said: “If we are going to have self-driving cars, well maybe we should test them before putting them on the road rather than beta testing them live on the road

can't read the article (pay wall) or find the royal society speech online - yet ?

but Demis's credentials are impeccable https://www.youtube.com/watch?v=SZpINa6UvpQ

edit - when i have 90 minutes too spare https://www.youtube.com/watch?v=zYfzux7JKHE

 

[Freakbro]

I don't think he has a problem with self driving cars, just he is saying that they shouldn't be "tested" on the open roads and shouldn't mix with human drivers until they are totally ready.

In a perfect world I guess that would be possible.

 

[sesevans]

this article is getting some attention in the uk


can't read the article (pay wall) or find the royal society speech online - yet ?

but Demis's credentials are impeccable https://www.youtube.com/watch?v=SZpINa6UvpQ

edit - when i have 90 minutes too spare https://www.youtube.com/watch?v=zYfzux7JKHE

And the irony of his position is that Google bought Deep Mind approximately 4 years ago for a reported £ 400 million in 2014. Alphabet, Google's parent company, is the owner of Waymo.

But this is very Google....having one part of the business taking a view that does not necessarily conform to the view of another part of their business. They run their Company like a giant debating society! Lots of strongly held, often divergent views. Some have said it is an extension of the way Larry and Sergey experienced their Stanford University days.

Despite this seeming contradiction, I believe Deep Mind will prove to be one of Google's greatest investments. Deep Mind is an Artificial Intelligence company. A significant collaboration has already taken place with Deep Mind providing AI to control temperatures/air conditioning in Google's giant datacentres. DeepMind says its AI can make the cooling units in Google's data centres 40% more efficient, ultimately cutting the data centres' overall electricity consumption by 15%. I have seen estimates that this savings along could amount to hundreds of millions of dollars annually, more than paying back Google's initial investment.

Deep Mind has also contributed significantly to Waymo's operations too. And as Google has become increasingly reliant on AI in their cloud computing business, Google Cloud's AI offering has become leading edge. As users of Google services, we have also seen Deep Mind's AI contributions in many areas such as GMail, Google Shopping, Google Maps, Google's latest Duplex technology, Google Lens, etc.

 

[sesevans]

Slightly off topic about Uber, I noticed today that Lyft, Uber's main US based competitor, is now claiming it holds a 35% market share in the US ride hailing market, its highest ever, and up from 20% 18 months ago.

Market share numbers tend to vary, depending on your source but Lyft growth appears on a firm footing.

https://www.cnbc.com/2018/05/14/lyft-market-share-051418-bosa-sf.html

Separately, here in the UK, Uber continues to fight to reverse the decision to ban it from London. Today they appointed a new UK chief executive:

https://uk.reuters.com/article/uk-u...-battles-to-keep-london-licence-idUKKCN1IF1A9

 

[sesevans]

According to the Wall St Journal, and before the fatal crash near San Francisco of a Tesla Model X vehicle in March, Tesla considered adding eye tracking and steering wheel sensors to their Autopilot system to add more warnings to inattentive drivers but rejected these ideas due to concerns over effectiveness and cost.

 

[sesevans]

We must be close to witnessing the launch of the first AV commercial ride hailing service. Today Waymo issued a 41 page manual for police officers which included how to break in and disable the Waymo self driving vehicle mainly for the purpose of immobilising the vehicle but secondly to protect the Waymo autonomous technology.

Why issue and help train the police if you are not about to launch?

https://www.cnbc.com/2018/05/14/alp...ow-to-break-in-disable-self-driving-cars.html