Autonomous Cars

[roon]

I believe by 2020 just about all cars will be convertible with some work,

Just in time for global warming.

[Big Tea]

I believe by 2020 just about all cars will be convertible with some work,

Just in time for global warming.

Don't go there, I just had a warning for off topic

[theblackangus]

Many vehicles today seem to have the 'bare bones' already included in them.
I have ordered a low end car for delivery Sept 1 and it has the 'Camera' for anti collision and lane departure sign reader rear sensors cruse etc. (Its Automatic) High end cars have had most of this for some time.

I think just about everything has electric power steering today which can be converted, braking system could be modular if the 'brain' is there

I believe by 2020 just about all cars will be convertible with some work, soon after with plug in modules.

I dont think its anywhere near that simple.
For real autonomous driving that works in all environments (not just some idea areas) you will need lidar, radar, vision (full 360), and possibly ultra sonic (for close quarters collision) and maybe other items, not to mention high resolution GPS which isn't cheap yet. (Getting cheaper tho!)
Autonomous driving is a correlation challenge, you need multiple types of sensors (and makes) to ensure you are not
confused by a condition in the environment or a bug as autonomous driving is a life and death scenario.

Uber disabled the "Fail Safe" in the accident they caused because the car was stopping so much it became to inconvenient to make sense. (I read it would hit a false positive less than every 5 minutes that would cause the car to stop).

And then to "fail safe" w/o having it be a total pita, you will need multiple copies of these on redundant circuits so if a camera, or lidar fails or gets too dirty there are backups. Also you need different models of each sensor on the vehicle to ensure you are not hitting the same bug across various sensors at the same time.

I build and fly DIY autonomous drones, and even with something so simple there have been alot of hardware and software bugs that caused issues. Today most drones contain different makes/models of inertial sensors, GPS, Gyros, etc to ensure they all agree with each other and if not invoke the fail safe because a bug or condition that it cannot handle has been hit.

As a question to the group for full 95% of environments autonomy:
What sensors do you believe will be needed?
Would you feel safe with only one sensor of each type knowing there could be a firmware bug?
What type of software validation should be done for bugs and testing that would make you feel comfortable?

Personally I think there needs to be regulations and 3rd party validation of code along the lines of what the aero space industry does.

[Just_a_fan]

Synchronized braking should reduce/eliminate traffic waves.

That would make it enormously safer, but I am old school, and was schooled at the time that brakes are for emergencies and miscalculations.

Brakes are for slowing, gears are for going.

The days of brakes made from cheese are long gone. Use them, that's what they are there for.

[Just_a_fan]

However that would ruin it for us adrenaline junky drivers. The John Force's of the world would hate it.

John Force would still be able to drive on his quarter mile of straight tarmac with no one coming the other way. AVs won't spell the end of motorsport (or drag racing).

[Just_a_fan]

Many vehicles today seem to have the 'bare bones' already included in them.
I have ordered a low end car for delivery Sept 1 and it has the 'Camera' for anti collision and lane departure sign reader rear sensors cruse etc. (Its Automatic) High end cars have had most of this for some time.

I think just about everything has electric power steering today which can be converted, braking system could be modular if the 'brain' is there

I believe by 2020 just about all cars will be convertible with some work, soon after with plug in modules.

I dont think its anywhere near that simple. [rest of quote removed for ease of forum navigation]

I find the whole idea that AVs must be as close to perfect as possible quite interesting. It seems to me that so long as an AV is as good as the average human driver then we're ok. Anything better than that improves the overall situation.

Most humans are pretty rubbish at driving, to be honest. Using phones, chatting to passengers, fiddling with the radio, putting on make up, watching the fine bit of on the sidewalk, thinking about dinner/work/the kids etc. Any AV that can keep its view on the road and deal as best it can with what's going on is better than all of those drivers.

After that, AV systems will improve and so overall he roads will improve too.

Remember that for most human drivers, the car is just a means to an end. They don't take pride or pleasure in driving, it's just a chore required in order to get from A to B in a convenient manner.

[strad]

@ Just-A-Fan My reply was to AJI as noted in my post.
.
Settling for AVs to only being as good as humans kind of defeats the idea of them being in the interest of safety.

[theblackangus]

I find the whole idea that AVs must be as close to perfect as possible quite interesting. It seems to me that so long as an AV is as good as the average human driver then we're ok. Anything better than that improves the overall situation.

I don't think what I have specified is close to perfect mostly because software is written by humans. That is to get acceptable results across a wide spectrum of driving environments in the winter/snowing, dark, rain, hail, fog, etc and standard speeds w/o outlawing everything else from the roads.

Most humans are pretty rubbish at driving, to be honest. Using phones, chatting to passengers, fiddling with the radio, putting on make up, watching the fine bit of on the sidewalk, thinking about dinner/work/the kids etc. Any AV that can keep its view on the road and deal as best it can with what's going on is better than all of those drivers.

Right and just one sensor cannot currently do the job that you have stated with any real safety. Do you want to stop driving on a rainy day when mud keeps getting on your camera? The car is no good if it fails safely every mile.

[Just_a_fan]

@ Just-A-Fan My reply was to AJI as noted in my post.
.
Settling for AVs to only being as good as humans kind of defeats the idea of them being in the interest of safety.

As I said, if AVs are as good as humans to start with, they will improve and systems improve. Humans won't, to be honest.

[Just_a_fan]

Right and just one sensor cannot currently do the job that you have stated with any real safety. Do you want to stop driving on a rainy day when mud keeps getting on your camera? The car is no good if it fails safely every mile.

Humans basically rely on one sensor - their vision. What do we do when mud/rain/snow gets in front of our eyes? We wipe the windscreen. Why can't AVs have wipers in front of their sensors (or the equivalent)?

The reality is that an AV can have a multitude of sensors, monitor them all at the same time, talk to neighbouring AVs and borrow their sensor data, make use of road-side sensor data etc. It works for humans, affter all - we get info from active traffic signs, we get info from other vehicles (brake lights), we can also use other sensors (assuming the radio is off). But humans get bored, tired, distracted, lazy, don't care. We're also pretty poor at doing two or more things at the same time.

The only issue slowing AVs is implementation. That and politics, of course.

[roon]

Humans basically rely on one sensor - their vision.

Hmm. Really should be tapping abilities of the clairvoyants, channelers, mystics, fortune tellers, palm readers, and psychics. This would eliminate most of the issues with current AV technologies. Predictive avoidance. Can't get in an accident if you know it's gonna happen. Huge opportunity missed there by the tech companies there, a real blind spot. A spoon bender mounted in the glove box would help allay the doubts of the more critical passengers as to the miracles bestowed of the sacred software.

[theblackangus]

Right and just one sensor cannot currently do the job that you have stated with any real safety. Do you want to stop driving on a rainy day when mud keeps getting on your camera? The car is no good if it fails safely every mile.

Humans basically rely on one sensor - their vision. What do we do when mud/rain/snow gets in front of our eyes? We wipe the windscreen. Why can't AVs have wipers in front of their sensors (or the equivalent)?

We work way better with snow, rain, mud, etc than current vision software does. That is why most all AV companies are using a combination of Lidar, Radar, and vision systems.

The reality is that an AV can have a multitude of sensors, monitor them all at the same time, talk to neighbouring AVs and borrow their sensor data, make use of road-side sensor data etc. It works for humans, affter all - we get info from active traffic signs, we get info from other vehicles (brake lights), we can also use other sensors (assuming the radio is off). But humans get bored, tired, distracted, lazy, don't care. We're also pretty poor at doing two or more things at the same time.

This was my point... to be safe AV's have to have alot of sensors and input to ensure they are not plagued with false positives.

The only issue slowing AVs is implementation. That and politics, of course.

Saying the only issue that that is slowing AV adoption is implementation.... Well thats the key to everything. Ideas are a dime a dozen, but getting it to work well is the gold.
Getting that to work as well as us in 95% of the driving scenarios (the minimum bar) hasn't happened yet, not even close.

Are you saying that you think AV software today can use a simple camera and be as good humans?
Im somewhat confused, as I was replying to a post that said cars have most of the sensors they need today, and I was pointing out that they didn't to get a reasonable level of safety.

[Big Tea]

Synchronized braking should reduce/eliminate traffic waves.

That would make it enormously safer, but I am old school, and was schooled at the time that brakes are for emergencies and miscalculations.

Brakes are for slowing, gears are for going.

The days of brakes made from cheese are long gone. Use them, that's what they are there for.

I disagree. I see no reason to floor it until you are 6 inches behind the car in front then stand on the break.
You go just as quickly as the car in front, which is all you can ever do, and get a few fractions extra time if you need it to react. Try it you may deprecate it. A much more relaxed way to get places. But I don't mean hold up the traffic either. There are many who drive that way but you would not know if you were not in the car with them, unless you ask them how long tyres last.

[Big Tea]

Right and just one sensor cannot currently do the job that you have stated with any real safety. Do you want to stop driving on a rainy day when mud keeps getting on your camera? The car is no good if it fails safely every mile.

Humans basically rely on one sensor - their vision. What do we do when mud/rain/snow gets in front of our eyes? We wipe the windscreen. Why can't AVs have wipers in front of their sensors (or the equivalent)?

We work way better with snow, rain, mud, etc than current vision software does. That is why most all AV companies are using a combination of Lidar, Radar, and vision systems.

The reality is that an AV can have a multitude of sensors, monitor them all at the same time, talk to neighbouring AVs and borrow their sensor data, make use of road-side sensor data etc. It works for humans, affter all - we get info from active traffic signs, we get info from other vehicles (brake lights), we can also use other sensors (assuming the radio is off). But humans get bored, tired, distracted, lazy, don't care. We're also pretty poor at doing two or more things at the same time.

This was my point... to be safe AV's have to have alot of sensors and input to ensure they are not plagued with false positives.

The only issue slowing AVs is implementation. That and politics, of course.

Saying the only issue that that is slowing AV adoption is implementation.... Well thats the key to everything. Ideas are a dime a dozen, but getting it to work well is the gold.
Getting that to work as well as us in 95% of the driving scenarios (the minimum bar) hasn't happened yet, not even close.

Are you saying that you think AV software today can use a simple camera and be as good humans?
Im somewhat confused, as I was replying to a post that said cars have most of the sensors they need today, and I was pointing out that they didn't to get a reasonable level of safety.

What I would like to pick up on here is variables. Not just between one sensor (person) and another, but the same on at different times and at different parts of the life curve. Yes you can have a very good very safe driver, but if the kids are playing hell in the back and the wife is standing on the roadside where he was not expecting her to be picked up, the 'loop' will not be the same as optimum.

The other huge advantage is human drivers never know what the other vehicles are about to do. We know what they are supposed to do and what we think the are gong to do. The autonomous car will know because they will be in constant exchange with the other vehicles. No guesswork.

There will still be accidents, there always will, but far less and probably far less severe than now.

[Just_a_fan]

That would make it enormously safer, but I am old school, and was schooled at the time that brakes are for emergencies and miscalculations.

Brakes are for slowing, gears are for going.

The days of brakes made from cheese are long gone. Use them, that's what they are there for.

I disagree. I see no reason to floor it until you are 6 inches behind the car in front then stand on the break.
You go just as quickly as the car in front, which is all you can ever do, and get a few fractions extra time if you need it to react. Try it you may deprecate it. A much more relaxed way to get places. But I don't mean hold up the traffic either. There are many who drive that way but you would not know if you were not in the car with them, unless you ask them how long tyres last.

Ah, yes, I see what you mean now. I thought you were rehashing the argument I've had in the past about "using the gears to slow down and only use the brakes as a last resort/to finish slowing". The sort of stuff my dad did when he were a lad, really, because the brakes were rubbish back in the day.

I drive as you suggest, unless I have a bit of open road and I'm in the car on my own. Then I might have a bit of fun and then the brakes do get used somewhat more firmly...