Toyota previews self driving car(bbc.co.uk)
bbc.co.uk
Toyota previews self driving car
http://www.bbc.co.uk/news/technology-20910769
7 comments
One issue with automated trains is when something inevitably goes wrong, if it happens in a heavy trafficked system like in NYC you might have dozens to hundreds of people dead compared to a couple of people with a driver-less car incident, which will make it much bigger news, which will mean a much bigger backlash.
And safety experts will go on the news and explain how the train incident was one issue among billions of safely-moved people and it was really a human error that was the root cause and the automated trains are much, much safer than human drivers by the statistics and they'll be completely right, but people won't care because omg robots.
And safety experts will go on the news and explain how the train incident was one issue among billions of safely-moved people and it was really a human error that was the root cause and the automated trains are much, much safer than human drivers by the statistics and they'll be completely right, but people won't care because omg robots.
One issue with automated trains is when something inevitably goes wrong, if it happens in a heavy trafficked system like in NYC you might have dozens to hundreds of people dead
Can you give an example of a subway accident (in NYC or another US system) that lead to that many dead? A head-on collision on the WMATA system resulted in only 9 dead: http://en.wikipedia.org/wiki/June_2009_Washington_Metro_trai...
From this list of NYC subway accidents, the worst involved an accident in 1991 that killed 5 passengers: http://www.nycsubway.org/wiki/Subway_FAQ:_Accidents
Can you give an example of a subway accident (in NYC or another US system) that lead to that many dead? A head-on collision on the WMATA system resulted in only 9 dead: http://en.wikipedia.org/wiki/June_2009_Washington_Metro_trai...
From this list of NYC subway accidents, the worst involved an accident in 1991 that killed 5 passengers: http://www.nycsubway.org/wiki/Subway_FAQ:_Accidents
Nope I can't. The numbers I randomly made up are certainly on the high side, but they were made on the assumption that along with automation there would be other changes like increasing the average travelling speed of the trains by a significant amount and having increased loads due to better scheduling algorithms.
I don't doubt there will be some of that. One can hope their will be fewer issues than on human operated trains, and the issues will not result in the loss of life or even the same scale of injury. People ride in elevators after all.
But more importantly is that if it is 'safe enough' such that a critical mass of people use it, then the anti-tech antibodies won't matter. Some people refuse to fly, others to drink tap water, but the system is 'safe enough' that its self sustaining and can tolerate lack of participation by these folks.
But more importantly is that if it is 'safe enough' such that a critical mass of people use it, then the anti-tech antibodies won't matter. Some people refuse to fly, others to drink tap water, but the system is 'safe enough' that its self sustaining and can tolerate lack of participation by these folks.
This actually exists now [1]. All that is left is to scale it up. They could probably even make versions of the pods that'll work with existing tracks.
[1] http://news.bbc.co.uk/2/hi/uk_news/7148731.stm
[1] http://news.bbc.co.uk/2/hi/uk_news/7148731.stm
Fabulous! I didn't know they had gotten the funding to go ahead with that system. I really hope it transforms the way people think about Heathrow, in so doing it will generate something one can point to without any use of the word 'imagine if there existed ...'
Why not combine both? This proposal recently won a design contest, the tech already exists from maglev systems. http://www.youtube.com/watch?v=9wLokWkUdHM
From the description the BBC gave, it seems they're behind the curve:
"A series of optical beacons on the roadside can detect the positions of pedestrians and obstacles, and relay information to the prototype about whether a traffic light is red or green, as part of ITS. The car can also independently monitor pedestrians' positions."
What roads are going to be set up with special optical beacons so the car knows where things are? If that was possible, we would have had self-driving cars a decade ago.
What roads are going to be set up with special optical beacons so the car knows where things are? If that was possible, we would have had self-driving cars a decade ago.
It sounds like they just tested with those beacons as a backup or alternate system.
Supposedly the Google driverless car uses a Velodyne lidar system that, after some processing, provides a 3d point-cloud describing the scene around the car.
I assume that Toyota is using a similar system. I am pretty sure that lidar system is the whole key to making driverless cars feasible because it provides so much realtime data about the entire area around the car. http://velodynelidar.com/lidar/hdlpressroom/How%20Googles%20...
http://velodynelidar.com/lidar/hdlproducts/hdl64e.aspx
After reading about how that lidar provides a detailed 3d map of the environment for many meters out, I'm not even sure its really a particularly interesting or cutting edge AI project.
I was thinking something similar might be true for natural language understanding: the more data points your map of natural language usage has, the less general your intelligence problem becomes.
Supposedly the Google driverless car uses a Velodyne lidar system that, after some processing, provides a 3d point-cloud describing the scene around the car.
I assume that Toyota is using a similar system. I am pretty sure that lidar system is the whole key to making driverless cars feasible because it provides so much realtime data about the entire area around the car. http://velodynelidar.com/lidar/hdlpressroom/How%20Googles%20...
http://velodynelidar.com/lidar/hdlproducts/hdl64e.aspx
After reading about how that lidar provides a detailed 3d map of the environment for many meters out, I'm not even sure its really a particularly interesting or cutting edge AI project.
I was thinking something similar might be true for natural language understanding: the more data points your map of natural language usage has, the less general your intelligence problem becomes.
An open problem that Google has admitted with their car automation project is a certain problem endemic to the northern part of this country: Snow and Ice.
Admittedly, ice is rather difficult in any condition. However, many human drivers drive safely on snow cover with minimal problems. Accidents are higher during those times, but traffic still keeps flowing. Google-car-AI cannot handle these events at all (at this time).
Admittedly, ice is rather difficult in any condition. However, many human drivers drive safely on snow cover with minimal problems. Accidents are higher during those times, but traffic still keeps flowing. Google-car-AI cannot handle these events at all (at this time).
If we're at a point where having to take manual control of a robot car during icey weather is a problem, then I'd say we are in a pretty good place.
I'm not sure I agree. If the only time perole's driving skills are used is during the most treacherous time to be driving, that will be scary.
I already avoid driving during the first snow storm of the season because, after six months, people have inevitably forgotten that ice is slippery (seriously, it a frightening and dangerous to be on the roads during the first storm!). Now factor in the fact nobody has driven at all during the last six months and it will be exceedingly messy.
I already avoid driving during the first snow storm of the season because, after six months, people have inevitably forgotten that ice is slippery (seriously, it a frightening and dangerous to be on the roads during the first storm!). Now factor in the fact nobody has driven at all during the last six months and it will be exceedingly messy.
The Twin Cities (MN) two year ago had about 400 accidents during the big blizzard at the start of the year. I hesitate to think what would have happened if those folks hadn't had to drive all summer.
Yeah, that seems like an easy problem to solve, with both technology and regulation. Self-driving cars would only get certified for driving conditions they can handle, so driving below, say, 3°C or with ice or snow on the street would not be allowed if a car can’t handle ice or snow. (I assume that in the foreseeable future all self-driving cars will also have to support some sort of manual driving.) The first condition can even be enforced automatically. When self-driving cars get better those conditions can be relaxed.
I think I'd trust Google more than a car company to make self driving cars. That right there is one of the problems though, which self driving car do you trust the most?
That of the car company, of course. I bet its car would beat any car Google built as far as passive safety goes, and don't see how any software could compensate for that. Even if software could, I do not think (but am less sure of it) Google's self-driving car software is that much better than those of car companies, if at all.
In other words: “car companies have learned and struggled for a few years here figuring out how to make a decent car. Google is not going to just figure this out. They’re not going to just walk in.”
In other words: “car companies have learned and struggled for a few years here figuring out how to make a decent car. Google is not going to just figure this out. They’re not going to just walk in.”
Issues of liability would be interesting to deal with...
Car companies TODAY have been shipping en masse large number of self-parking and automatic accident avoiding cars and have decades of experience in building and shipping reliable control systems.
Google can't even produce bug-free software, make high quality hardware or even ship anything at any decent scale.
Google can't even produce bug-free software, make high quality hardware or even ship anything at any decent scale.
> self-parking and automatic accident avoiding
Unfortunately these are really simple systems relative to safely driving on the road without supervision. If we're talking about the same things, then self-parking is just a software for recognising the front/end of some vehicle parked on the side and where is the good parking space - in some models also a distance monitor for automatic stopping. It's not that difficult to implement and all the tricky conditions still have to be checked by people - for example my auto-parking car will completely ignore people in its way, anything on the street, will abort if it's on a slope down, will not check for lots of issues that you're supposed to correct manually.
Collision avoidance is for now a simple break when some object is getting close fast. Only a handful of cars implement something more as far as I know, since actively changing the course is actually a rather large challenge to do it right...
I have to disagree with "Google can't even produce bug-free software" too. There's a big difference between programming a web application and doing something that's close to safety critical software. It's a completely different set of skills and requires a specific approach in general. Even then... regarding bug-free at scale, I think they're doing fairly well so far.
Unfortunately these are really simple systems relative to safely driving on the road without supervision. If we're talking about the same things, then self-parking is just a software for recognising the front/end of some vehicle parked on the side and where is the good parking space - in some models also a distance monitor for automatic stopping. It's not that difficult to implement and all the tricky conditions still have to be checked by people - for example my auto-parking car will completely ignore people in its way, anything on the street, will abort if it's on a slope down, will not check for lots of issues that you're supposed to correct manually.
Collision avoidance is for now a simple break when some object is getting close fast. Only a handful of cars implement something more as far as I know, since actively changing the course is actually a rather large challenge to do it right...
I have to disagree with "Google can't even produce bug-free software" too. There's a big difference between programming a web application and doing something that's close to safety critical software. It's a completely different set of skills and requires a specific approach in general. Even then... regarding bug-free at scale, I think they're doing fairly well so far.
Not that I'm arguing with your assessment of car companies' experiences, but self-parking and autonomous driving are completely different beasts. I wouldn't be surprised, if the first company that creates a viable driverless car isn't one of the incumbent car companies. As Tesla have shown with electric car.
I'm sure they could produce (mostly) bug-free software if the cost-benefit ratio made sense. The problem with trying to ensure a web application is bug-free is that your incremental spend is probably greater than the incremental revenue. Self-driving cars are a different story. They will get sued when the cars make mistakes.
The article is unclear, but it sounds like Google is not involved in this.
Why isn't Google selling this technology to car companies? It seems like an excellent second revenue stream.
Why isn't Google selling this technology to car companies? It seems like an excellent second revenue stream.
Why should they buy it? I think a lot of people in the tech community are under the impression that this is something Google has invented and that this will be a new cash cow. The reality is car companies have been working on this technology for a while, with many "intelligent" features already existing in higher end cars.
A friend from grad school was working on this at Toyota about 6 years ago, so I can confirm this didn't somehow take the car companies by surprise.
Google seems to be ahead of the curve in this tech. I'd love to be proven wrong, have any links?
I've been reading about this randomly over the years so I don't have any links in mind, these are off Wikipedia.
http://www.thenational.ae/business/technology/an-automated-a... http://www.wired.com/autopia/2008/01/gm-says-driverl/ http://www.audiusa.com/us/brand/en/tools/news/pool/2010/07/n... http://media.vw.com/newsrelease.do;jsessionid=89FFC8CECBE997... http://www.motorauthority.com/news/1068584_2013-mercedes-ben... http://media.cadillac.com/media/us/en/cadillac/news.detail.h... http://www.bbc.co.uk/news/technology-19829906
http://www.thenational.ae/business/technology/an-automated-a... http://www.wired.com/autopia/2008/01/gm-says-driverl/ http://www.audiusa.com/us/brand/en/tools/news/pool/2010/07/n... http://media.vw.com/newsrelease.do;jsessionid=89FFC8CECBE997... http://www.motorauthority.com/news/1068584_2013-mercedes-ben... http://media.cadillac.com/media/us/en/cadillac/news.detail.h... http://www.bbc.co.uk/news/technology-19829906
Also: http://en.wikipedia.org/wiki/EUREKA_Prometheus_Project.
A mean time between human interventions of 9 km may not seem something to brag about, but this project ended before Google was founded aka 10 Moore's generations ago. I think Mercedes has continued working on driverless cars ever since.
A mean time between human interventions of 9 km may not seem something to brag about, but this project ended before Google was founded aka 10 Moore's generations ago. I think Mercedes has continued working on driverless cars ever since.
Volvo are planning to ship their first self driving cars in 2014.
http://blogs.wsj.com/drivers-seat/2012/12/03/volvo-plans-to-...
http://blogs.wsj.com/drivers-seat/2012/12/03/volvo-plans-to-...
How much time to mainstream acceptance? I'd say ten years.
Tricky one. I think there will be initial positivity. We're getting to that stage now. Lawmakers are happy to embrace it at the moment. The first accidents will result in a major setback, though. Difficult to say exactly how much people will freak out when the first casualties happen in a self-driving car. Hopefully by that time there will be enough data to strongly demonstrate that even though self-driving cars can still crash, they can also be much safer than non-autonomous cars.
Not in Europe.
Still shifting manually.
To be in control.
The European mood is shifting (no pun intended). I see an increasing number of models being offered with automatic gearboxes, even at the lower end of the price scale.
Sales numbers are probably available somewhere, but, for a quick sample, I took a cursory look at a local used car site - out of 5058 cars listed from 2010 onwards, some 716 are automatic. Nearly 15% is pretty good for something that used to be much harder to come by.
Sales numbers are probably available somewhere, but, for a quick sample, I took a cursory look at a local used car site - out of 5058 cars listed from 2010 onwards, some 716 are automatic. Nearly 15% is pretty good for something that used to be much harder to come by.
I think in general we're at a point now where everything is going to come increasingly faster than we expect. The singularity is near after all :)
I agree! Round number for sure!
"The car can also communicate with other vehicles"
This topic is interesting in quite a few aspects. Firstly, it's something I thought of long ago, but due to implementation details (ECM is a closed, obfuscated PITA).
The biggest ideas revolving around this com device is how it processes the communication. Is the communication an authenticated protocol? Being an amateur radio operator and a fellow hacker, I can think of plenty of things to do regarding this device, both good and bad. Does it take "suggestions" from other devices? If so, do these suggestions get stored in a sort of a log to prove action your AI takes is the fault of another?
And there's the security involved in the local processor. Many in this community are turned off to the idea of secure computing. In this case however, could lead to a bad user fake-signing messages to other cars. This is an open question, due to many secure implementations successfully hacked (Nintendo DS, PS2, PS3, Blu-Ray, various media stores).
I'm sure there is a way to make a car-puter that interfaces with other vehicles for a mesh network. However, I would believe that this network would have to be one of those close to being mathematically provable languages. And I would not think that TOYOTA would be capable of this.
This topic is interesting in quite a few aspects. Firstly, it's something I thought of long ago, but due to implementation details (ECM is a closed, obfuscated PITA).
The biggest ideas revolving around this com device is how it processes the communication. Is the communication an authenticated protocol? Being an amateur radio operator and a fellow hacker, I can think of plenty of things to do regarding this device, both good and bad. Does it take "suggestions" from other devices? If so, do these suggestions get stored in a sort of a log to prove action your AI takes is the fault of another?
And there's the security involved in the local processor. Many in this community are turned off to the idea of secure computing. In this case however, could lead to a bad user fake-signing messages to other cars. This is an open question, due to many secure implementations successfully hacked (Nintendo DS, PS2, PS3, Blu-Ray, various media stores).
I'm sure there is a way to make a car-puter that interfaces with other vehicles for a mesh network. However, I would believe that this network would have to be one of those close to being mathematically provable languages. And I would not think that TOYOTA would be capable of this.
Car 2 Car communication is already ready:
https://events.ccc.de/congress/2012/Fahrplan/events/5095.en....
https://events.ccc.de/congress/2012/Fahrplan/events/5095.en....
You might be interested in the SARTRE platooning project currently being tested in the EU: http://www.sartre-project.eu/en/Sidor/default.aspx
Food for thought: the only reason traffic jams happen is because humans are imperfect creatures that can't maintain a uniform speed (barring a total-closure collision or some other abnormal event).
Imagine the impact on humans if traffic jams just vanished. Engineers have known for a while that even in the heaviest congestion (think 405 or 580) it's usually only a mistake by a driver that will completely stop traffic. I wouldn't mind an auto-drive that was just speed control, enforced in situations of congestion. Rather than one person braking too hard and completely stopping a mile of 880, everybody gets programmed to travel at a fixed speed.
A lot of things to work out there, but if vehicles worked together with the road to maintain their own speed, traffic would be light years better.
Imagine the impact on humans if traffic jams just vanished. Engineers have known for a while that even in the heaviest congestion (think 405 or 580) it's usually only a mistake by a driver that will completely stop traffic. I wouldn't mind an auto-drive that was just speed control, enforced in situations of congestion. Rather than one person braking too hard and completely stopping a mile of 880, everybody gets programmed to travel at a fixed speed.
A lot of things to work out there, but if vehicles worked together with the road to maintain their own speed, traffic would be light years better.
It's really a matter of slow and inconsistent reaction to changes in speed, not the change in speed itself. The delayed reaction of one car causes it to decelerate more then they would have had to, then for the car behind the impact is greater and so on. Simulations suggest that as little as 20% of vehicles having computer-controlled reactions (very fast) would just about eliminate these shockwave traffic jams.
Related, see #6 in this article: http://www.nytimes.com/interactive/2012/06/03/magazine/innov...
Related, see #6 in this article: http://www.nytimes.com/interactive/2012/06/03/magazine/innov...
This can be solved on the cheap with pace cars, as used in several European countries.
Consider the overall effectiveness vs engineering effort compared to say 'smarter mass transit'. Imagine how useful it would be to have a mass transit system which took advantage of the extreme efficiency of rolling on rails, with vehicle based switching system [1]. We don't do that today because a) trains can't change lanes if something bad is going to happen, and b) the system is optimized for large trains going on the same route.
But autonomously driving a train is a much easier problem than autonomously driving a car. And if you created a system where vehicles could self-select to enter a station then you could create mass transit with 'call on demand' vehicles which only stopped at the pickup and drop off stations. This eliminates the scheduling friction which has empty trains running at 2AM and overly full trains running at 5PM.
Building a system like that is well within our current engineering capabilities, would save energy (these things are electric not gas powered) and achieve high ridership by being very nearly as convenient as a car without the hassle of driving in traffic.
[1] Traditional rail switches use arms on the ground to configure a rail switch to move a car on to one track or another, in a vehicle based system the vehicle configures its wheels to take one path or the other, this allows the track to be 'passive' and not need a central switching authority.