Autonomous Driving Features in Cars 2020

[TorqueDog]

Originally Posted by GreenLantern2814

I know how these systems work, and I use them all the time - they are for use in clear conditions, as they do not adjust for them.

We are talking about autonomous driving systems that aren't available to consumers yet, and you aren't the first person to work at a car dealership -- not that it gives one any authority over the subject.

[mrkajz44]

I do find it interesting how this though process works:

1) We all agree that computers are vastly better than humans at taking in a ton of inputs, computing them, and spitting out appropriate outputs
2) We also all agree that winter driving requires a driver to consider far more inputs than normal driving (how slippery is the road, will I start to skid if I hit the brakes, etc.)

But after putting these two together, the conclusion is that human drivers are better than computers at winter driving. I don't get it.

[Fuzz]

Originally Posted by mrkajz44

I do find it interesting how this though process works:

1) We all agree that computers are vastly better than humans at taking in a ton of inputs, computing them, and spitting out appropriate outputs
2) We also all agree that winter driving requires a driver to consider far more inputs than normal driving (how slippery is the road, will I start to skid if I hit the brakes, etc.)

But after putting these two together, the conclusion is that human drivers are better than computers at winter driving. I don't get it.

Because there are some things computers are not very good at. Like, you and I can look at a dog, and say "that's a dog". An algorithm can look at it, and have no idea what it is. You can see a cyclist, and see that they saw you, and understand each others intentions. Computers are not good at that. They'll get there, but we have a ways to go.



For winter driving, most of these systems rely on lines on the road to see where they are. Without it, they become useless, whereas humans can make interpretations of inputs. It's not that computers can't process all the inputs, it's that computers need to be programmed to interpret them. They have to identify features in an image properly which is much tougher when everything is covered in snow.

[powderjunkie]

Originally Posted by mrkajz44

I do find it interesting how this though process works:

1) We all agree that computers are vastly better than humans at taking in a ton of inputs, computing them, and spitting out appropriate outputs
2) We also all agree that winter driving requires a driver to consider far more inputs than normal driving (how slippery is the road, will I start to skid if I hit the brakes, etc.)

But after putting these two together, the conclusion is that human drivers are better than computers at winter driving. I don't get it.

The Inuit have 50 different words for snow for a reason.

Determining those inputs is a big challenge in itself, let alone transforming them to outputs that achieve anything resembling efficient travel. Acceleration ain't hard, it's the opposite that's probably going to be the problem. Will computerized risk tolerances co-exist with human drivers in an open environment?

I'm really curious how much specific effort manufacturers are putting into these relatively niche situations? There may be a lot of affluent consumers who live in places that get severe winters, but it's still a fairly small piece of the pie.

[JohnnyB]

Originally Posted by Fuzz

Because there are some things computers are not very good at. Like, you and I can look at a dog, and say "that's a dog". An algorithm can look at it, and have no idea what it is. You can see a cyclist, and see that they saw you, and understand each others intentions. Computers are not good at that. They'll get there, but we have a ways to go.



For winter driving, most of these systems rely on lines on the road to see where they are. Without it, they become useless, whereas humans can make interpretations of inputs. It's not that computers can't process all the inputs, it's that computers need to be programmed to interpret them. They have to identify features in an image properly which is much tougher when everything is covered in snow.

Not that I disagree with the overall thrust of where you're going, but computers are now better at image recognition than humans are. Also, they get there not by being programmed to understand what each thing is, but by learning from large data sets with millions of examples. There are issues with the type and variety of sensors used as well as the volume of data and quality of data though, and that seems to fit with what I take your point to be.

[TorqueDog]

I will happily admit that visual recognition of roadways in a scenario where road markings are obscured will pose an issue. Hell, most of the time humans aren't even driving in the actual lanes when the roadways are so snow covered that the lines are not visible. Maybe cities will need to add road infrastructure that uses some form of standardized beaconing system to give the car a digital view of the road's layout. Imagine that... digital lane closures.

Where computers outshine humans is monitoring hundreds of data points at once and being able to manipulate outputs at a level far more granular than humans are capable of (or availed of in some instances).

Case in point, stability control systems. I don't care how good of a driver you are, you cannot decide to apply only the rear far-side brake caliper at 46% pressure while maintaining a steering angle of precisely 47 degrees given the surface friction of the roadway that has been calculated based off of wheel rotation via ABS sensors as well as the inputs received from the yaw sensor in the stability control system... then milliseconds later adjust the brake pressure to a different caliper entirely and steering angle to exactly the right values to mesh with new information from that plethora of sensors and resulting calculations. All the while the person behind the wheel is probably doing things counter-productive to the goal of keeping the vehicle from skidding into a guardrail.

The reason I am very confident in driverless technology being able to cope with winter driving is because we are removing the erratic, unpredictable variable behind the wheel. Sure there will be hurdles, but these systems are becoming incredibly powerful.

[mrkajz44]

Originally Posted by TorqueDog

I will happily admit that visual recognition of roadways in a scenario where road markings are obscured will pose an issue. Hell, most of the time humans aren't even driving in the actual lanes when the roadways are so snow covered that the lines are not visible. Maybe cities will need to add road infrastructure that uses some form of standardized beaconing system to give the car a digital view of the road's layout. Imagine that... digital lane closures.

Where computers outshine humans is monitoring hundreds of data points at once and being able to manipulate outputs at a level far more granular than humans are capable of (or availed of in some instances).

Case in point, stability control systems. I don't care how good of a driver you are, you cannot decide to apply only the rear far-side brake caliper at 46% pressure while maintaining a steering angle of precisely 47 degrees given the surface friction of the roadway that has been calculated based off of wheel rotation via ABS sensors as well as the inputs received from the yaw sensor in the stability control system... then milliseconds later adjust the brake pressure to a different caliper entirely and steering angle to exactly the right values to mesh with new information from that plethora of sensors and resulting calculations. All the while the person behind the wheel is probably doing things counter-productive to the goal of keeping the vehicle from skidding into a guardrail.

The reason I am very confident in driverless technology being able to cope with winter driving is because we are removing the erratic, unpredictable variable behind the wheel. Sure there will be hurdles, but these systems are becoming incredibly powerful.

This is where I am living too. Even if computers struggle with certain aspects of winter driving (no clear lane markings for example), they are still much better than humans in so many other aspects that computers will clearly be the better winter drivers very soon.

[Fuzz]

I don't disagree that they will get there. My only point is they aren't there yet, and it will be awhile. I suspect in the next 2-5 years we see FSD working in ideal conditions. I suspect it will be 10 years before they get Canadian winter driving figured out.

[Fuzz]

Hrmm, lets check in on roborace, the entirely autonomous racing series...

https://twitter.com/user/status/1321800383505657856


"And the start has not gone to plan" is a small understatement.

[powderjunkie]

Originally Posted by Fuzz

Hrmm, lets check in on roborace, the entirely autonomous racing series...

https://twitter.com/user/status/1321800383505657856


"And the start has not gone to plan" is a small understatement.

LOL.

It will be cool if we can eventually see race cars like Adrian Newey's concept on the Gran Turismo series, where the limiting factor is the human anatomy dealing with the G forces (and ability to handle to car IRL)

[TorqueDog]

Originally Posted by Fuzz

I don't disagree that they will get there. My only point is they aren't there yet, and it will be awhile.

I don't think anyone disagrees with this point.

[Fuzz]

Originally Posted by TorqueDog

I don't think anyone disagrees with this point.

Well then, now that we are all in agreement, I guess we can all go home. Wait, I'm already home. Forever.

[powderjunkie]

I suppose the the important thing to remember with these sorts of things is how overly optimistic timeline estimates tend to be. It's not necessarily about what is/isn't possible, but when.

I bet it's at least 20 years before we see much consumer-level winter full AV tech.

[Azure]

From a safety perspective, even without FSD, I wonder if we can get to a place where vehicles can make decisions to avoid accidents.

Like running a red light, intersections, head ons, etc.

[Flamenspiel]

Originally Posted by mrkajz44

I do find it interesting how this though process works:

1) We all agree that computers are vastly better than humans at taking in a ton of inputs, computing them, and spitting out appropriate outputs
2) We also all agree that winter driving requires a driver to consider far more inputs than normal driving (how slippery is the road, will I start to skid if I hit the brakes, etc.)

But after putting these two together, the conclusion is that human drivers are better than computers at winter driving. I don't get it.

It depends, there are skilled winter drivers and unskilled ones. It takes a lot of practice.

Ultimately computers are designed and programmed by humans, so you are not really removing human error from the equation. Then there is deliberate sabotage, will the car companies be providing daily patching and anti-virus signatures.

[TorqueDog]

Originally Posted by Azure

From a safety perspective, even without FSD, I wonder if we can get to a place where vehicles can make decisions to avoid accidents.

Like running a red light, intersections, head ons, etc.

The Moral Machine is a perfect representation of the dilemma in programming AI to make accident avoidance decisions that are left to humans.

[Enoch Root]

In some ways, I think it is unrealistic to expect self-driving vehicles to function independently. There are just so many random variables, as a quick for instance: trafiic lights are out due to power failure, cops are manually directing traffic, etc.

The answer to these types of issues, IMO, is to have a comprehensive system that is monitoring ALL traffic, and the system communicates with individual vehicles.

Unfortunately, a system like that is a LONG ways away. And it would suck because it would probably mean the end of personal driving. And I, for one, LIKE driving.

[Fuzz]

The only reasonable path is to have vehicles be able to handle all situations, which is the really tricky part, because then you have to teach a computer to think like a human, which is obviously not currently possible, and I don't think anyone sees it happening for a long time. So without that, you are left programming all foreseeable situations, which still leaves the unforeseen ones.



Having centralized systems that cars rely on is a non-starter, because they could go down, or be jammed etc. They can be used to supplement information, which will be really useful, but the car still needs to be able to reason out all situations. It may not have to do that optimally for everything, but we will expect them to do it safely.

[JohnnyB]

Originally Posted by Fuzz

The only reasonable path is to have vehicles be able to handle all situations, which is the really tricky part, because then you have to teach a computer to think like a human, which is obviously not currently possible, and I don't think anyone sees it happening for a long time. So without that, you are left programming all foreseeable situations, which still leaves the unforeseen ones.



Having centralized systems that cars rely on is a non-starter, because they could go down, or be jammed etc. They can be used to supplement information, which will be really useful, but the car still needs to be able to reason out all situations. It may not have to do that optimally for everything, but we will expect them to do it safely.

Nobody would be doing that. That's not what people do. If each situation had to be programmed for, it would be an impossible task to have autonomous vehicles, but systems with such vast amount of input data as autonomous vehicles depend upon neural nets to learn what works and how to act in novel circumstances without having each scenario planned for.

[Fuzz]

Tesla Inc (TSLA.O) was sued on Wednesday in a proposed class action accusing Elon Musk's electric car company of misleading the public by falsely advertising its Autopilot and Full Self-Driving features.
The complaint accused Tesla and Musk of having since 2016 deceptively advertised the technology as fully functioning or "just around the corner" despite knowing that the technology did not work or was nonexistent, and made vehicles unsafe.

Briggs Matsko, the named plaintiff, said Tesla did this to "generate excitement" about its vehicles, attract investments, boost sales, avoid bankruptcy, drive up its stock price and become a "dominant player" in electric vehicles.
"Tesla has yet to produce anything even remotely approaching a fully self-driving car," Matsko said.

The lawsuit filed in federal court in San Francisco seeks unspecified damages for people who since 2016 bought or leased Tesla vehicles with Autopilot, Enhanced Autopilot and Full Self-Driving features.

https://www.reuters.com/business/aut...ms-2022-09-14/

Tesla released a new beta that was supposed to be a huge improvement, but the videos I've watched shows it still struggles. It's a long ways from being reliable. My brother says his has tried to blow through stop signs and run red lights. Just like real people!