Autonomous vehicles

dbs · 07-01-2018, 08:32 PM

Originally posted by veato

Given the standard of driving I experienced this morning on the way to work the sooner we have autonomous vehicles the better

Hi Veato
what do you feel has changed with todays drivers from only 20yrs ago ?

Dave

I feel habits have changed too

PDL · 07-01-2018, 08:42 PM

Originally posted by Racer X 69

I know about driving in the winter.

--snip

The entire state of Wyoming was shut down for a week in mid April 2013. When they finally opened it up it was slow because thousands of trucks got on the road at the same time. The surface was again, graded and sanded, no salt.

But it was real slick. Took all day to go from Rawlins to Cheyenne, and they actually shut it down for a few hours when I was just getting to Laramie. I stopped in to a place and found a nice restaraunt and had lunch. When I hit the road after it opened it up yet again it looked like this.

--snip

Been there, done that. I was raised in Laramie, and I know the "Snow Chi Minh" trail well. I have been in a stopped pickup full of Coca Cola stopped just out side of Arlington and being blown side ways nearly off the road by the wind. Ahh - those were the days.

dbs · 07-01-2018, 08:55 PM

When we get autonomous vehicles I think they will be electric of some kind and not petrol/diesel ( probably to expensive if any left by then anyway )
..I don't really see it being commonplace in my lifetime/

Dave

LensBeginner · 07-02-2018, 12:11 AM

Originally posted by Racer X 69

I'm thinking that work on this is moving toward GPS location rather than Earth based controls. Although signal lights may become "smart" and be fitted to communicate with autonomous vehicles. And of course the vehicles themselves having on board detection devices to aid keeping them moving or not moving safely. And likely a wi-fi link to a central or regional traffic control center.

The communications systems will probably also be able to feed real time conditions in urban areas with cameras at intersections, etc. So when there is a utility crew blocking a lane the system will know and avoid the work zone.

@PDL 's objection was that GPS doesn't work in a whiteout.
Apart from that, I agree - also considering that by "GPS" we now refer to more than the GPS constellation proper (GLONASS, BeiDou, Galileo...)

MarkJerling · 07-02-2018, 03:31 PM

Most modern cars with radar, lidar and cameras all round see better than any human can. This is because they need not only use the visible light spectrum like humans. Some, like the Mercedes system, see perfectly fine through fog. What these systems are not that good with yet, is being able to work out whether the thing they're seeing is a child on a bicycle, a dog, a car or a power pole. So, much time and effort goes into programming the algorithms to ensure that the car can work out what it's seeing - correctly.

Liney · 07-02-2018, 06:05 PM - **1 Like**

I agree with you Mark, but the trouble is with computers is they need to be given specific instructions what do to with the information. I'm sure there was another thread recently about artificial intelligence, and this is one of the cases where AI would be required.

I read some of the reports on recently reported incident where an autonomously driven vehicle collided with a pedestrian who was pushing a bike across the road. A lot of the reports pointed the finger at the "driver" (or rather the person who was sitting in the driving seat monitoring the cars progress) as they were not looking at the road at the time and could have prevented the collision. Personally I've had experience of being distracted by looking down at the centre console for something and looking up to find I'm feet from impacting a rather large red deer, so in this case I don't think the "driver" was any more guilty than the rest of us.

But the one thing that struck me about the incident was that the human pushing the bike should have been able to see the vehicle coming yet still decided to cross the road at that point.

Autonomous vehicles need to be programmed (in the strictest sense of the word) to drive defensively. That programming needs to track every other moving object in the vicinity and if there is the potential for an impact either slow or stop the car or taking avoiding action.

As we still have car incidents every day I would suggest we've failed to program human drivers to do it, so what hope have we got with machines?

Racer X 69 · 07-02-2018, 06:17 PM

Originally posted by Liney

But the one thing that struck me about the incident was that the human pushing the bike should have been able to see the vehicle coming yet still decided to cross the road at that point.

I had the same reaction when I saw the video.

Why did the cyclist ignore the oncoming vehicle, and avoid crossing in front of it?

MarkJerling · 07-02-2018, 06:44 PM

Originally posted by Liney

I agree with you Mark, but the trouble is with computers is they need to be given specific instructions what do to with the information. I'm sure there was another thread recently about artificial intelligence, and this is one of the cases where AI would be required.

I read some of the reports on recently reported incident where an autonomously driven vehicle collided with a pedestrian who was pushing a bike across the road. A lot of the reports pointed the finger at the "driver" (or rather the person who was sitting in the driving seat monitoring the cars progress) as they were not looking at the road at the time and could have prevented the collision. Personally I've had experience of being distracted by looking down at the centre console for something and looking up to find I'm feet from impacting a rather large red deer, so in this case I don't think the "driver" was any more guilty than the rest of us.

But the one thing that struck me about the incident was that the human pushing the bike should have been able to see the vehicle coming yet still decided to cross the road at that point.

Autonomous vehicles need to be programmed (in the strictest sense of the word) to drive defensively. That programming needs to track every other moving object in the vicinity and if there is the potential for an impact either slow or stop the car or taking avoiding action.

As we still have car incidents every day I would suggest we've failed to program human drivers to do it, so what hope have we got with machines?

As I said, it's the algorithms to enable the machine to identify what it's seeing that is the issue. Much work is being done with that. I have no doubt that machines will soon enable us to choose to read a book on the way to work, should we wish to. Personally, I'd prefer to drive myself!

The autonomous vehicle / pedestrian with bicycle crash is interesting. The car did have a driver. He was not paying attention and no-one knows if the outcome would have been different if he was.

Here's the crash: The lady later died - so avoid watching if you find this material too sensitive.

Youtu.be

MarkJerling · 07-02-2018, 06:51 PM - **1 Like**

Originally posted by Racer X 69

I had the same reaction when I saw the video.

Why did the cyclist ignore the oncoming vehicle, and avoid crossing in front of it?

I suspect she did not hear the very quiet Volvo, but it could be that she expected the car to slow or stop for her. A very unfortunate event.

dbs · 07-02-2018, 07:53 PM - **2 Likes**

Originally posted by Liney

I agree with you Mark, but the trouble is with computers is they need to be given specific instructions what do to with the information. I'm sure there was another thread recently about artificial intelligence, and this is one of the cases where AI would be required.

I read some of the reports on recently reported incident where an autonomously driven vehicle collided with a pedestrian who was pushing a bike across the road. A lot of the reports pointed the finger at the "driver" (or rather the person who was sitting in the driving seat monitoring the cars progress) as they were not looking at the road at the time and could have prevented the collision. Personally I've had experience of being distracted by looking down at the centre console for something and looking up to find I'm feet from impacting a rather large red deer, so in this case I don't think the "driver" was any more guilty than the rest of us.

But the one thing that struck me about the incident was that the human pushing the bike should have been able to see the vehicle coming yet still decided to cross the road at that point.

Autonomous vehicles need to be programmed (in the strictest sense of the word) to drive defensively. That programming needs to track every other moving object in the vicinity and if there is the potential for an impact either slow or stop the car or taking avoiding action.

As we still have car incidents every day I would suggest we've failed to program human drivers to do it, so what hope have we got with machines?

Hi Liney
All the AI protections have to be in place but then no vehicle would ever move because there's always a percieved danger.

Dave

PDL · 07-02-2018, 08:51 PM

I have read on several sites, news sites, internet tech sites etc. that not only was the "driver" not looking out the windshield and did not have their hands on the steering wheel, that Uber had purposely disabled the Volvo LIDAR system. So the car was relying on Uber systems only and some of the sites stated that the "driver" was streaming something from Hulu.

Now a camera based system that can not recognize a pedestrian pushing a bicycle in front of it is in my opinion not much of a system. How could the system not see the person and after a few milliseconds know that it was going to hit it. Especially when the "autonomous" part of this scenario is that the car will miss big things in the road. And this was on a clear night in Arizona, no rain, no snow, no fog, no dust storms. If this is state of the art, then we are in for a very long wait. I would really like to see autonomous vehicles come about as I like to zone out and listen to music when I am on the bus but I am starting to think that I will be long gone before it happens.

MarkJerling · 07-02-2018, 09:12 PM - **1 Like**

Originally posted by PDL

I have read on several sites, news sites, internet tech sites etc. that not only was the "driver" not looking out the windshield and did not have their hands on the steering wheel, that Uber had purposely disabled the Volvo LIDAR system. So the car was relying on Uber systems only and some of the sites stated that the "driver" was streaming something from Hulu.

Now a camera based system that can not recognize a pedestrian pushing a bicycle in front of it is in my opinion not much of a system. How could the system not see the person and after a few milliseconds know that it was going to hit it. Especially when the "autonomous" part of this scenario is that the car will miss big things in the road. And this was on a clear night in Arizona, no rain, no snow, no fog, no dust storms. If this is state of the art, then we are in for a very long wait. I would really like to see autonomous vehicles come about as I like to zone out and listen to music when I am on the bus but I am starting to think that I will be long gone before it happens.

According to the data extracted from the system, the car 'saw' the pedestrian.

"According to data obtained from the self-driving system, the system first registered radar and LIDAR observations of the pedestrian about 6 seconds before impact, when the vehicle was traveling at 43 mph. As the vehicle and pedestrian paths converged, the self-driving system software classified the pedestrian as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path. At 1.3 seconds before impact, the self-driving system determined that an emergency braking maneuver was needed to mitigate a collision. According to Uber, emergency braking maneuvers are not enabled while the vehicle is under computer control, to reduce the potential for erratic vehicle behavior. The vehicle operator is relied on to intervene and take action. "

As this is an experimental system, and not a fully autonomous vehicle, the driver was supposed to intervene. She did not, as she was not looking at the road at the time of the collision.

reh321 · 07-02-2018, 09:26 PM

I wonder how well an autonomous vehicle could follow some of the rural roads around here; I have a terrible time doing it at night, so if its super vision could enable it to do better - than might be a desirable thing.

Liney · 07-03-2018, 03:35 AM

They ran an autonomous vehicle trial on the local freeway last year, using cardboard cutouts of "obstacles" to see how it would react. The car successfully "saw" and stopped for other car shapes and people shapes, but ran straight through a kangaroo shape....

photoptimist · 07-03-2018, 06:22 AM - **1 Like**

Originally posted by MarkJerling

According to the data extracted from the system, the car 'saw' the pedestrian.

"According to data obtained from the self-driving system, the system first registered radar and LIDAR observations of the pedestrian about 6 seconds before impact, when the vehicle was traveling at 43 mph. As the vehicle and pedestrian paths converged, the self-driving system software classified the pedestrian as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path. At 1.3 seconds before impact, the self-driving system determined that an emergency braking maneuver was needed to mitigate a collision. According to Uber, emergency braking maneuvers are not enabled while the vehicle is under computer control, to reduce the potential for erratic vehicle behavior. The vehicle operator is relied on to intervene and take action. "

As this is an experimental system, and not a fully autonomous vehicle, the driver was supposed to intervene. She did not, as she was not looking at the road at the time of the collision.

This kind of testing -- with the computer watching the road but not fully controlling the car -- is common during development. It collects a lot of data on how the sensors and control algorithms would likely react to varying real-world conditions without relying on them if those systems are not ready to take full control.

There was clearly a serious bit of human negligence behind this tragedy. Certainly the driver should have been paying more attention. But it's also possible that the manager(s) or engineer(s) overseeing the testing were negligent in not clearly instructing the driver about what the computer was controlling or ignoring.

This event and some of the accidents involving Tesla's mis-named "Autopilot" highlight one of the challenges to developing these kinds of systems. Partial automation (SAE levels 2 and 3) may be worse than no automation at all. If the car's systems are good, the driver quickly becomes habituated to the car handling the driving. But if the car's systems are not good enough, the driver won't be paying any attention to the road when the system fails to recognize a dangerous situation.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Some incredibly FAST vehicles!	Dewman	General Talk	6	06-02-2018 10:58 AM
First driving experiences....what vehicles, who with, interesting stories..	lesmore49	General Talk	31	05-17-2018 10:17 AM
Project 52 Project 52-6-16-SUBJECT-VEHICLES	noelcmn	Weekly Photo Challenges	11	06-09-2014 11:02 AM
Vehicles in my street	hooppjs	Post Your Photos!	8	01-29-2008 06:33 PM
From moving vehicles (2 images)	hamidlmt	Post Your Photos!	3	12-03-2007 10:05 AM

07-01-2018, 08:32 PM	#46
dbs Pentaxian Join Date: Jun 2010 Location: Clare Valley S A Photos: Albums Posts: 7,563	Originally posted by veato Given the standard of driving I experienced this morning on the way to work the sooner we have autonomous vehicles the better Hi Veato what do you feel has changed with todays drivers from only 20yrs ago ? Dave I feel habits have changed too

07-01-2018, 08:42 PM	#47
PDL Pentaxian Join Date: Oct 2006 Location: PNW USA Photos: Gallery Posts: 2,128	Originally posted by Racer X 69 I know about driving in the winter. --snip The entire state of Wyoming was shut down for a week in mid April 2013. When they finally opened it up it was slow because thousands of trucks got on the road at the same time. The surface was again, graded and sanded, no salt. But it was real slick. Took all day to go from Rawlins to Cheyenne, and they actually shut it down for a few hours when I was just getting to Laramie. I stopped in to a place and found a nice restaraunt and had lunch. When I hit the road after it opened it up yet again it looked like this. --snip Been there, done that. I was raised in Laramie, and I know the "Snow Chi Minh" trail well. I have been in a stopped pickup full of Coca Cola stopped just out side of Arlington and being blown side ways nearly off the road by the wind. Ahh - those were the days.

07-01-2018, 08:55 PM	#48
dbs Pentaxian Join Date: Jun 2010 Location: Clare Valley S A Photos: Albums Posts: 7,563	When we get autonomous vehicles I think they will be electric of some kind and not petrol/diesel ( probably to expensive if any left by then anyway ) ..I don't really see it being commonplace in my lifetime/ Dave

07-02-2018, 12:11 AM	#49
LensBeginner Veteran Member Join Date: Sep 2014 Photos: Albums Posts: 4,696	Originally posted by Racer X 69 I'm thinking that work on this is moving toward GPS location rather than Earth based controls. Although signal lights may become "smart" and be fitted to communicate with autonomous vehicles. And of course the vehicles themselves having on board detection devices to aid keeping them moving or not moving safely. And likely a wi-fi link to a central or regional traffic control center. The communications systems will probably also be able to feed real time conditions in urban areas with cameras at intersections, etc. So when there is a utility crew blocking a lane the system will know and avoid the work zone. @PDL 's objection was that GPS doesn't work in a whiteout. Apart from that, I agree - also considering that by "GPS" we now refer to more than the GPS constellation proper (GLONASS, BeiDou, Galileo...)

07-02-2018, 03:31 PM	#50
MarkJerling Moderator Site Supporter Join Date: May 2012 Location: Wairarapa, New Zealand Photos: Gallery \| Albums Posts: 20,406	Most modern cars with radar, lidar and cameras all round see better than any human can. This is because they need not only use the visible light spectrum like humans. Some, like the Mercedes system, see perfectly fine through fog. What these systems are not that good with yet, is being able to work out whether the thing they're seeing is a child on a bicycle, a dog, a car or a power pole. So, much time and effort goes into programming the algorithms to ensure that the car can work out what it's seeing - correctly.

07-02-2018, 06:05 PM - 1 Like	#51
Liney Veteran Member Join Date: Apr 2013 Posts: 1,237	I agree with you Mark, but the trouble is with computers is they need to be given specific instructions what do to with the information. I'm sure there was another thread recently about artificial intelligence, and this is one of the cases where AI would be required. I read some of the reports on recently reported incident where an autonomously driven vehicle collided with a pedestrian who was pushing a bike across the road. A lot of the reports pointed the finger at the "driver" (or rather the person who was sitting in the driving seat monitoring the cars progress) as they were not looking at the road at the time and could have prevented the collision. Personally I've had experience of being distracted by looking down at the centre console for something and looking up to find I'm feet from impacting a rather large red deer, so in this case I don't think the "driver" was any more guilty than the rest of us. But the one thing that struck me about the incident was that the human pushing the bike should have been able to see the vehicle coming yet still decided to cross the road at that point. Autonomous vehicles need to be programmed (in the strictest sense of the word) to drive defensively. That programming needs to track every other moving object in the vicinity and if there is the potential for an impact either slow or stop the car or taking avoiding action. As we still have car incidents every day I would suggest we've failed to program human drivers to do it, so what hope have we got with machines?

07-02-2018, 06:17 PM	#52
Racer X 69 Moderator Man With A Camera Join Date: Jun 2010 Location: The Great Pacific Northwet, in the Land Between Canada and Mexico Photos: Gallery \| Albums Posts: 28,068	Originally posted by Liney But the one thing that struck me about the incident was that the human pushing the bike should have been able to see the vehicle coming yet still decided to cross the road at that point. I had the same reaction when I saw the video. Why did the cyclist ignore the oncoming vehicle, and avoid crossing in front of it?

07-02-2018, 06:51 PM - 1 Like	#54
MarkJerling Moderator Site Supporter Join Date: May 2012 Location: Wairarapa, New Zealand Photos: Gallery \| Albums Posts: 20,406	Originally posted by Racer X 69 I had the same reaction when I saw the video. Why did the cyclist ignore the oncoming vehicle, and avoid crossing in front of it? I suspect she did not hear the very quiet Volvo, but it could be that she expected the car to slow or stop for her. A very unfortunate event.

07-02-2018, 07:53 PM - 2 Likes	#55
dbs Pentaxian Join Date: Jun 2010 Location: Clare Valley S A Photos: Albums Posts: 7,563	Originally posted by Liney I agree with you Mark, but the trouble is with computers is they need to be given specific instructions what do to with the information. I'm sure there was another thread recently about artificial intelligence, and this is one of the cases where AI would be required. I read some of the reports on recently reported incident where an autonomously driven vehicle collided with a pedestrian who was pushing a bike across the road. A lot of the reports pointed the finger at the "driver" (or rather the person who was sitting in the driving seat monitoring the cars progress) as they were not looking at the road at the time and could have prevented the collision. Personally I've had experience of being distracted by looking down at the centre console for something and looking up to find I'm feet from impacting a rather large red deer, so in this case I don't think the "driver" was any more guilty than the rest of us. But the one thing that struck me about the incident was that the human pushing the bike should have been able to see the vehicle coming yet still decided to cross the road at that point. Autonomous vehicles need to be programmed (in the strictest sense of the word) to drive defensively. That programming needs to track every other moving object in the vicinity and if there is the potential for an impact either slow or stop the car or taking avoiding action. As we still have car incidents every day I would suggest we've failed to program human drivers to do it, so what hope have we got with machines? Hi Liney All the AI protections have to be in place but then no vehicle would ever move because there's always a percieved danger. Dave

07-02-2018, 08:51 PM	#56
PDL Pentaxian Join Date: Oct 2006 Location: PNW USA Photos: Gallery Posts: 2,128	I have read on several sites, news sites, internet tech sites etc. that not only was the "driver" not looking out the windshield and did not have their hands on the steering wheel, that Uber had purposely disabled the Volvo LIDAR system. So the car was relying on Uber systems only and some of the sites stated that the "driver" was streaming something from Hulu. Now a camera based system that can not recognize a pedestrian pushing a bicycle in front of it is in my opinion not much of a system. How could the system not see the person and after a few milliseconds know that it was going to hit it. Especially when the "autonomous" part of this scenario is that the car will miss big things in the road. And this was on a clear night in Arizona, no rain, no snow, no fog, no dust storms. If this is state of the art, then we are in for a very long wait. I would really like to see autonomous vehicles come about as I like to zone out and listen to music when I am on the bus but I am starting to think that I will be long gone before it happens.

07-02-2018, 09:12 PM - 1 Like	#57
MarkJerling Moderator Site Supporter Join Date: May 2012 Location: Wairarapa, New Zealand Photos: Gallery \| Albums Posts: 20,406	Originally posted by PDL I have read on several sites, news sites, internet tech sites etc. that not only was the "driver" not looking out the windshield and did not have their hands on the steering wheel, that Uber had purposely disabled the Volvo LIDAR system. So the car was relying on Uber systems only and some of the sites stated that the "driver" was streaming something from Hulu. Now a camera based system that can not recognize a pedestrian pushing a bicycle in front of it is in my opinion not much of a system. How could the system not see the person and after a few milliseconds know that it was going to hit it. Especially when the "autonomous" part of this scenario is that the car will miss big things in the road. And this was on a clear night in Arizona, no rain, no snow, no fog, no dust storms. If this is state of the art, then we are in for a very long wait. I would really like to see autonomous vehicles come about as I like to zone out and listen to music when I am on the bus but I am starting to think that I will be long gone before it happens. According to the data extracted from the system, the car 'saw' the pedestrian. "According to data obtained from the self-driving system, the system first registered radar and LIDAR observations of the pedestrian about 6 seconds before impact, when the vehicle was traveling at 43 mph. As the vehicle and pedestrian paths converged, the self-driving system software classified the pedestrian as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path. At 1.3 seconds before impact, the self-driving system determined that an emergency braking maneuver was needed to mitigate a collision. According to Uber, emergency braking maneuvers are not enabled while the vehicle is under computer control, to reduce the potential for erratic vehicle behavior. The vehicle operator is relied on to intervene and take action. " As this is an experimental system, and not a fully autonomous vehicle, the driver was supposed to intervene. She did not, as she was not looking at the road at the time of the collision.

07-02-2018, 09:26 PM	#58
reh321 Pentaxian Join Date: Nov 2014 Location: South Bend, IN, USA Photos: Gallery \| Albums Posts: 23,180	I wonder how well an autonomous vehicle could follow some of the rural roads around here; I have a terrible time doing it at night, so if its super vision could enable it to do better - than might be a desirable thing.

07-03-2018, 03:35 AM	#59
Liney Veteran Member Join Date: Apr 2013 Posts: 1,237	They ran an autonomous vehicle trial on the local freeway last year, using cardboard cutouts of "obstacles" to see how it would react. The car successfully "saw" and stopped for other car shapes and people shapes, but ran straight through a kangaroo shape....

07-03-2018, 06:22 AM - 1 Like	#60
photoptimist Pentaxian Join Date: Jul 2016 Photos: Albums Posts: 5,126	Originally posted by MarkJerling According to the data extracted from the system, the car 'saw' the pedestrian. "According to data obtained from the self-driving system, the system first registered radar and LIDAR observations of the pedestrian about 6 seconds before impact, when the vehicle was traveling at 43 mph. As the vehicle and pedestrian paths converged, the self-driving system software classified the pedestrian as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path. At 1.3 seconds before impact, the self-driving system determined that an emergency braking maneuver was needed to mitigate a collision. According to Uber, emergency braking maneuvers are not enabled while the vehicle is under computer control, to reduce the potential for erratic vehicle behavior. The vehicle operator is relied on to intervene and take action. " As this is an experimental system, and not a fully autonomous vehicle, the driver was supposed to intervene. She did not, as she was not looking at the road at the time of the collision. This kind of testing -- with the computer watching the road but not fully controlling the car -- is common during development. It collects a lot of data on how the sensors and control algorithms would likely react to varying real-world conditions without relying on them if those systems are not ready to take full control. There was clearly a serious bit of human negligence behind this tragedy. Certainly the driver should have been paying more attention. But it's also possible that the manager(s) or engineer(s) overseeing the testing were negligent in not clearly instructing the driver about what the computer was controlling or ignoring. This event and some of the accidents involving Tesla's mis-named "Autopilot" highlight one of the challenges to developing these kinds of systems. Partial automation (SAE levels 2 and 3) may be worse than no automation at all. If the car's systems are good, the driver quickly becomes habituated to the car handling the driving. But if the car's systems are not good enough, the driver won't be paying any attention to the road when the system fails to recognize a dangerous situation.