Google Robot Cars

[maxx]

http://www.thedailybeast.com/blogs-a...s-car-sharing/
"...Not just one or two robots sneaking a few miles down a lonely country road late at night, but eight autonomous vehicles traveling more than 140,000 miles in the last year on everything from freeways to traffic-clogged downtown districts. I have viewed the route maps for several of these trips and can attest to the fact that Google’s car-bots have confidently traversed roads that would leave many drivers white-knuckled, from San Francisco’s pretzel-twisty Lombard Street to Big Sur’s narrow, cliff-hugging Highway One...."

[Jimaz]

http://www.thedailybeast.com/blogs-a...s-car-sharing/
"...Not just one or two robots sneaking a few miles down a lonely country road late at night, but eight autonomous vehicles traveling more than 140,000 miles in the last year on everything from freeways to traffic-clogged downtown districts. I have viewed the route maps for several of these trips and can attest to the fact that Google’s car-bots have confidently traversed roads that would leave many drivers white-knuckled, from San Francisco’s pretzel-twisty Lombard Street to Big Sur’s narrow, cliff-hugging Highway One...."

I have been closely watching automotive robotics ever since attending the first DARPA robotic Grand Challenge in 2005, and I know Sebastian Thrun, the entrepreneurial Stanford professor and Google distinguished software engineer who co-created Street View and now leads the team who put Google’s car-bots on the road. And like others, I sensed that something was up, from rumors of strange Google Street View vehicles seen on local highways to word that a particularly bright Stanford robotics Ph.D. student had disappeared into the Google campus. But nothing prepared me for the sheer magnitude of the surprise....

So was the surprisingly massive, worldwide Street View investment justified by its contribution to this carbot project?! Does carbot depend on Street View data to work its magic?

Bold. Very bold--not unusual for Google though. This is exciting.

rumors of strange Google Street View vehicles seen on local highways

To what does that refer?

[wolverine]

This technology should have been around alot sooner, not "5-10 years from now." We could have had a fully automated system in 2010. Problem is people love driving to much. They love to be behind the wheel which has made a fully automated system difficult to market. Personally I think someone driving a 200 mile trip would rather nap or do work on a laptop. There's also the crowd that hates driving in cities. Such a system would cut out human error and make driving a whole lot safer.

[ndavies]

This technology should have been around alot sooner, not "5-10 years from now." We could have had a fully automated system in 2010. Problem is people love driving to much. They love to be behind the wheel which has made a fully automated system difficult to market. Personally I think someone driving a 200 mile trip would rather nap or do work on a laptop. There's also the crowd that hates driving in cities. Such a system would cut out human error and make driving a whole lot safer.

What makes you believe these systems were ready to be used before today?

Darpa has run a contest since 2004 for fully automated vehicles. The military is very interested in unmanned supply vehicles. In 2004 none of the vehicles were able to complete the course.

In 2005 the second contest was held this was the first year any of the vehicles successfully completed the course. In that year 5 of the 23 finalists finished the course.

In 2007 they moved the competition to an simulated urban course. In this case the vehcles had to interact with each other. 6 cars finished the course in 2007, with hte winner posting an average speed of 14 miles an hour over thecourse.

All of the entrees in these contests were multimillion dollar custom develloped vehicles. They all contained technology that would be cost prohibitive for mass production.

If it's taking the highly funded military this long to get automated vehicles on the road, There's not much chance of us getting it in our daily drivers.

[Jimaz]

This is fascinating.

What should a stationary carbot do if some [[possibly drunken) human-driven vehicle approaches, threatening collision? Should it take evasive action in that situation? Why not?

It's a 2-D TCAS problem with the additional constraint of no option to pull up or descend.

[wolverine]

What makes you believe these systems were ready to be used before today?

Darpa has run a contest since 2004 for fully automated vehicles. The military is very interested in unmanned supply vehicles. In 2004 none of the vehicles were able to complete the course.

In 2005 the second contest was held this was the first year any of the vehicles successfully completed the course. In that year 5 of the 23 finalists finished the course.

In 2007 they moved the competition to an simulated urban course. In this case the vehcles had to interact with each other. 6 cars finished the course in 2007, with hte winner posting an average speed of 14 miles an hour over thecourse.

All of the entrees in these contests were multimillion dollar custom developed vehicles. They all contained technology that would be cost prohibitive for mass production.

If it's taking the highly funded military this long to get automated vehicles on the road, There's not much chance of us getting it in our daily drivers.

Easy answer. It has alot to do with research and funding. If we look at a wide range of technology and their development curves, you'd find them much steeper than AHS and driverless cars. This technology has been proven and ready to be marketed since the mid 90's. The reason it's garnered some attention is because a major company like Google is using it for streetview, which is a pretty smart idea IMO.

So if the military is "interested," I'm sure they took advantage of it over a decade ago, not in 2004. Why? Because google is using it in civilian vehicles to take pictures of streets everywhere.

[mjs]

This technology should have been around alot sooner, not "5-10 years from now." We could have had a fully automated system in 2010. Problem is people love driving to much. They love to be behind the wheel which has made a fully automated system difficult to market. Personally I think someone driving a 200 mile trip would rather nap or do work on a laptop. There's also the crowd that hates driving in cities. Such a system would cut out human error and make driving a whole lot safer.

I'm estatic about this and can't wait for it to become marketable. I'd nap during the morning commute and watch TV during the afternoon commute. However, marketing studies have shown that the problem is that people know that the computers they use fail from time to time so they fear giving full control to a computer. If the code fails, do you lose control or end up abandoned on the road? Does it know tree vs darting kid is different than tree vs darting squirrel? Manufacturers like Ford fear litigation and maintenance issues. It seems safe to assume Google followed specific maintenance and monitoring and had techs that understood the signs the vehicles were giving them. Would you or I? Some people struggle to even maintain regular oil changes.

Same issues with the cool fly by wire systems where the dash board and heavy steering column and its mechanical linkages are replaced with electronics and a center joystick. Driver seats further back with substantially less dashboard to slam into during an accident and increased interior space in a lighter, more fuel efficient vehicle. Awesome future.

[maxx]

This is fascinating.

What should a stationary carbot do if some [[possibly drunken) human-driven vehicle approaches, threatening collision? Should it take evasive action in that situation? Why not?

It's a 2-D TCAS problem with the additional constraint of no option to pull up or descend.

I want to know what happens to the passengers who are riding at 60mph inches from the next car when the carbot suddenly has to stop.

[ndavies]

Easy answer. It has alot to do with research and funding. If we look at a wide range of technology and their development curves, you'd find them much steeper than AHS and driverless cars. This technology has been proven and ready to be marketed since the mid 90's. The reason it's garnered some attention is because a major company like Google is using it for streetview, which is a pretty smart idea IMO.

So if the military is "interested," I'm sure they took advantage of it over a decade ago, not in 2004. Why? Because google is using it in civilian vehicles to take pictures of streets everywhere.

This technology has not been proven. Google has a limited prototype system. The military couldn't figure out how to do it, thus they held the Challenges.

It's the other fields that have been holding back automated vehicles. We have not had a powerful enough computing system that could cost effectively be put in a car. We have not had reliable enough sensor technology to detect the road and obstructions. We have not had the video object recognition algorithms needed to sense obstructions until recently.We have had the algorithms needed to drive, We have lacked a powerful enough computer system. Piloting a car down a typical street is a hugely complex task. It requires millions of decisions to be performed every second.

Computer technology has only reached a point where certain functions of automated vehicles could be cost effectively put into a car. The first of those systems are anti-lock brakes, stability control, adaptive cruise control and lane departure warning. These systems appear on the high priced vehicles as these cars can absorb the increased cost needed to put these systems in. More of these systems will be integrated into more cars as these systems reach a cost effective level.

Also all automated approaches developed before now required the roads to be modified so the car could find them. They required dedicated unobstructed lanes to operate correctly. They required grade seperation from all other vehicle and pedestrian traffic. All the cars in the system would have had to be controlled from a central location. There was no way to get enough computer technology into the car to allow them to be autonomous. The road modifications were designed to remove some of the sensing and computing requirements out of the car. Obviously this would have been outrageously expensive.

There are going to be huge people issues involved in getting automation into personal vehicles. Look at the outrage over Toyotas throttle peddle. Everyone's trying to blame the sudden accelleration problems on the mere fact that the mechanical throttle cable was replaced with computer control. Imagine what happens when an completely automated vehicle has sudden accelleration syndrome. Imagine hundreds of thousands of vehicles on the road with software bugs that could potentially run over children waiting for a school bus.

A friend of mine is currently working on steer by wire systems. This is the removal of the mechanical connection between the sterring wheel and the front wheels. The redundant safety control systems involved in this are huge. The steering wheel will be just like an arcade game steering wheel. It will send electrical signals to the power steering module which will then turn the direction the wheels are pointed.

What happens to drivers when these systems become reliable? Will the insurance companies price us out of being able to drive our own cars? I love to drive. Will I be once again priced out of something I truly love doing?

I have been developing vehicle control software for over 25 years. It's my day job. I am currently working on Vehicle to vehicle communication systems. One of the necessary technologies needed to simplify vehicle automated control. I can unequivocal state there has been no technology capable of driving a vehicle down a typical road before now. The military didn't have it and no vehicle manufacturer in the world had the technology capable of doing this. The sensors and computer technology have not been reliable or cheap enough to put this on a standard car.

IT definitely wasn't a lack of funding. Every atuo manufacturer and electronic module supplier has been trying to do it for at least the last 20 years. The US government has been dumping large amounts of military money at the project for a very long time. This is strictly a matter the technology not being anywhere near ready before now.

You should also see the multi dimensional sensor arrays on top of those DARPA automated vehicles. They're amazing pieces of technology I wish I could afford to get in my lab.

[ndavies]

I want to know what happens to the passengers who are riding at 60mph inches from the next car when the carbot suddenly has to stop.

As for passengers of the vehicle that's stopping. There would be a visual or auditory warning. You'd still need to wear seat belts. No worse than a current driver needing to do a panic stop.

As for surrounding cars. Vehicle to vehicle communications systems are going to be used to solve this issue. The robot car in front will warn the cars arround that it's stopping, why it's stopping and give a report on the road conditions it's currently seeing. So if it's trying to stop on ice it could warn the car behind it that there is ice ahead. The car behind would then know it had to brake earlier.

I can see automated cars allowing much higher densities of vehicles on the road and it should allow the cars to travel at much higher speeds.

[maxx]

Vehicle to vehicle communications systems are going to be used to solve this issue. The robot car in front will warn the cars arround that it's stopping, why it's stopping and give a report on the road conditions it's currently seeing. So if it's trying to stop on ice it could warn the car behind it that there is ice ahead. The car behind would then know it had to brake earlier.

Which is fine if most cars are carbots. Otherwise...

[ndavies]

Which is fine if most cars are carbots. Otherwise...

HUH, otherwise what? Carbots still have to have all signalling methods that are on other cars. You know, like brake lights and tail lights. They have to respond to stopped cars in front of them. They still have to keep a braking buffer distance in front of them. I see no difference between carbots and human driven cars in this regard. The whole challenge of a carbot is making it react to other obstructions [[like people and cars) in the road.

Adaptive cruise control with braking already far outperform a human in panic braking situations.