Lux Research: “The rush to proclaim driverless cars as the near future for personal transport is, simply, wrong”
18 August 2014
In a note following the 2014 Automated Vehicle Symposium, Lux Research said: “The rush to proclaim driverless cars as the near future for personal transport is, simply, wrong.” Lux participated in two separate sessions at the conference: the first around regional planning implications, and the second on personal vehicle automation commercialization.
Multiple times, we heard the phrase “But Google says...” uttered in protest, with various attendees citing its demonstration of autonomous vehicles and plans to pilot these vehicles in Northern California.
While many in the automotive value chain echo the conservative adoption rates from Lux’s studies, there was a vocal contingent, ranging from academics to government organizations, that used Google’s recent announcements as a counter argument. Upon further discussion, it was clear that as in any emerging technology, there will be ardent supporters that are predisposed to root for the adoption of the technology. The fact remains that Google’s demonstrations are technically impressive, and do represent a leap forward in autonomous technology; however, pilots and demonstrations are not the same as commercialization. The rush to proclaim driverless cars as the near future for personal transport is, simply, wrong.
While impressive, it is critical to understand that Google’s demonstrations are done in highly controlled environments, with consistent signage, clear weather, and precisely mapped roads (far beyond the level of detail for most locales). For example, Google has mapped the routes for its current vehicles down to 10 cm accuracy, and includes information around road markings—meaning in a place like Boston, where it's common to have leaves or snow on the ground, the car would not be able to operate safely. This makes machine vision one of the key areas to watch for innovation, because significant advances in both performance and cost must come when machines no longer rely on humans to see.
… human distraction and reaction pose a very difficult problem, and the solution may ultimately be an autonomous, driverless car that operates in limited-use cases, such as at low speed on predictable routes.
'and the solution may ultimately be an autonomous, driverless car that operates in limited-use cases, such as at low speed on predictable routes.'
Ultimately, huh?
Google, Nissan and umpteen universities beg to differ, and they have boots on the ground.
What academic or industrial resources has Lux deployed to reach these definitive conclusions?
Posted by: Davemart | 18 August 2014 at 02:53 AM
Driverless cars are coming and that's just the beginning. But has anyone here thought about what such things will ultimately mean to us?
https://www.youtube.com/watch?v=7Pq-S557XQU
Posted by: ai_vin | 18 August 2014 at 07:21 AM
If google are mapping down to 10cms and recording the positions of individual road markings, they are cheating a bit.
Humans can drive without detailed knowledge of the road layout, based on uncertain traffic markings and the ability to judge distance and just "know" where things are.
There is a lot of uncertainty in what people do and they take risks that it would be very hard to program for.
They continue driving when dazzled, or when the road edges and markings are obscured.
It sounds like google are trying to map an exact driving position at all times, in advance. But that is not the way humans drive, they can figure out the path at the last minute, as they drive along.
On the other hand, google can put much better 3d, vision, radar and lidar sensors on a car, while humans have just got short baseline colour stereo vision, and 600 million years of evolutionary development.
Humans can look at a scene, classify all objects and predict what each is likely to do next. (Post box - nothing, small boy - run out after ball, plastic bag - ignore) - without even knowing they are doing it.
This is very hard for a machine vision system, especially one running in real time on a road, day and night.
I think driving on a motorway should be easy enough, you could even let the driver nod off as long as you could either wake them up, or if this fails pull over to the side.
Another approach would be to slant the compensation rules so that if it looks like people won;t get money for getting run over by an AV, they will stay out of their ways. You will have recorded video of the drives, especially the last hour or so. Thus, if an accident occurs, it should be clear who is to blame: thus, it should be easy to calculate the payouts, as long as the legislation in in place in advance.
While you are developing AVs, you will kill a few people due to bugs and unforeseen circumstances. These people should get compensated (or their families should). What you do not want is people running out in front of AVs, hoping to get injured and sue Google. Full video retention should eliminate that problem.
@Al_Vin
In a best case, a car can drive you home while you are too tired or drunk to do so yourself.
Or you could read, surf or sleep while being driven.
It should be like having a driver 100% of the time.
Another idea is that you might have to pay premium insurance to use your car in Av mode (especially in towns) [ at least till they get the bugs out ]. Thus, you use AV mode as little as possible.
After they get the bugs out, and the machines become safer than us, you will have to pay more to drive it yourself.
Posted by: mahonj | 18 August 2014 at 08:20 AM
You don't need a 100% solution for driverless cars for them to be very useful. If they work on major roads and most places in cities that would be enough - you can drive yourself in rural areas or tricky unmarked roads.
When the car is about to reach the limit of automated operation (and it has a map of course and knows where it is going), then if you do not take over on request it can park itself and wait for you!!
It would suffice if you only have to drive the first and last few miles of each journey but the car does the rest.
Posted by: Peter | 18 August 2014 at 11:47 AM
'Fleets of self-driving lorries could be tested on UK roads as soon as next year, according to reports.
The technology allows a convoy of lorries to travel just a few feet from each other, with just the driver at the front in control.'
http://www.bbc.co.uk/news/technology-28834774
Posted by: Davemart | 18 August 2014 at 12:59 PM
"Google, Nissan and umpteen universities beg to differ, and they have boots on the ground.
What academic or industrial resources has Lux deployed to reach these definitive conclusions?"
Maybe the power of reason? These autonomous cars are far from being fully functional. It will take quite a while before it becomes viable.
Ultimately I think a network of driverless cars will be the norm with personal vehicles outright banned. (At least in towns/cities )
One just orders one, hops into it and then is taken to the exact destination.
Today we're at the robotic lawnmover level, which need guide wires but works.
Posted by: Zsolt Valaki | 15 September 2014 at 02:47 AM