The Unofficial "Elon Musk trying to "Save Everyone" from Themselves (except his Step-Sister)" Thread...

[CousinsCowgirl84]

Just now, tshile said:

That’s funny but not gonna work. The fact that you don’t even remember it falls in line with the rest of the issue and your track record. It is what it is. 

All you did was complain about how Tesla went about doing it and that you disagreed with it.

Ok, but that doesn’t show how it did real harm to autonomous driving, of which Tesla, partnering with mobile eye, was the first major proponent one.

 
The thing that’s done the most harm or autonomous driving is clickbait headlines claiming a autonomous car crashed when in fact numerous times after investigation was complete, the car was not even on autopilot.

[Cooked Crack]

 

[Captain Wiggles]

Musk gonna have to start branching out and diversifying his portfolio. Like what the My Pillow guy has done with his new My Slippers deal. 🤪

 

[The Evil Genius]

😅

 

 

[mistertim]

1 hour ago, CousinsCowgirl84 said:

All you did was complain about how Tesla went about doing it and that you disagreed with it.

Ok, but that doesn’t show how it did real harm to autonomous driving, of which Tesla, partnering with mobile eye, was the first major proponent one.

 
The thing that’s done the most harm or autonomous driving is clickbait headlines claiming a autonomous car crashed when in fact numerous times after investigation was complete, the car was not even on autopilot.

 

True autonomous driving (like we see in the movies) will never be a reality until the surrounding infrastructure is also as smart as the cars are. Trying to do all processing in the car is a fool's errand and creates single points of failure that are unacceptable in a life or death situation.

 

The car needs to be able to get data from and communicate with the surrounding environment in order to make the correct decisions.

 

Not doing it that way is like pilots flying planes into airports without any air traffic control systems, based purely on on-board radar and sight alone. Not happening.

[Hersh]

3 hours ago, tshile said:

Is that true though?

 

 

Don't know for certain but if $8 per gallon gas became the norm, I'm guessing it would change buying habits. Car companies would see this and adjust. 

 

[tshile]

19 minutes ago, mistertim said:

 

True autonomous driving (like we see in the movies) will never be a reality until the surrounding infrastructure is also as smart as the cars are. Trying to do all processing in the car is a fool's errand and creates single points of failure that are unacceptable in a life or death situation.

 

The car needs to be able to get data from and communicate with the surrounding environment in order to make the correct decisions.

 

Not doing it that way is like pilots flying planes into airports without any air traffic control systems, based purely on on-board radar and sight alone. Not happening.

I’ve always envisioned the end stage to involve some third party system that functions like air traffic control. It’ll do wonders for traffic but also make the system more reliable. 
 

im not sold it’s required for fully autonomous vehicles. It would make it easier, or more capable, reduce weird/odd behavior in edge cases, but I don’t think it’s required. I’m not exactly well suited to make that a hill I’m willing to die on, I just don’t think it’s required. They were able to create an autonomous vehicle and drive it across the country in the ‘80s. I think a lot more is possible without an extensive infrastructure. 
 

But really - I want to address the idea of “safe”. Not saying this is you, just a jumping off point on the topic. 

 

often when people are against something they’ll use that word and it’s a problem because it’s just loaded in what it means. There are people that would view any death/injury that can be blamed on an autonomous system as reason the system shouldn’t exist or be used. But that seems foolish to me. In this specific context, lots of people die in automobile accidents. If autonomous vehicles decreased that by 20%, that would be a win. And I think it’s more reasonable to think it would decrease them by significantly more, especially since almost all accidents are (by nature) from someone not following the rules. An autonomous system should follow the rules at a much higher rate. And I would expect that to net out very well against number of issues caused by autonomous vehicles. 
 

I expect the day autonomous vehicles become a real thing, we’ll have loads of people doing this. Pointing at each individual incident with outrage, completely ignoring the bigger picture. I expect that will also cause setbacks in the rollout. 
 

but it could almost eliminate deaths from dui, texting while driving, medical emergencies, exhausted/tired drivers, and stupid stuff like changing the radio station. 
 

but yeah the infrastructure would make it even better. But really I would expect that to make the overall system better for traffic routing and such, not make a cars individual movements better. And I’m not really sure you can offboard the important decision making - I don’t think you can depend on link latency there. I think you need immediate decision making. 

[tshile]

I also don’t like the airplane analogy. Air is a 3d space that relies on air traffic control to create and maintain lanes on the fly (though they obviously have rules about they do it) to avoid collisions. 
 

 Cars don’t that have. For all intents and purposes that’s a 2d space. And theoretically, in terms of safety, everything you need to know is right around you. Because the system was designed for humans, and it’s unreasonable to expect humans to make safe decisions without having immediate (and salient) aids to assist. That’s why we have the speed limit posted every so many feet even though it didn’t change; or, where appropriate, a sign that warns you that there’s another sign coming up that you need to be aware of but otherwise wouldn’t have enough time to safely process and act (like a stop sign just around the bend, that you can’t currently see.)

 

That said the 3rd party infrastructure could be anything and it’s possible we over exaggerate what it would look like. 
 

It’s entirely possible someone comes up with something you put in the paint used to line the roads that significantly improves the situation. Setting boundaries or signaling some important information (like hey this is now a work zone; speed limit 25mph; not 45 like your gps thinks it is; or in 250 feet it’s one way traffic with flag people controlling it so be prepared for that)

[mistertim]

9 minutes ago, tshile said:

I’ve always envisioned the end stage to involve some third party system that functions like air traffic control. It’ll do wonders for traffic but also make the system more reliable. 
 

im not sold it’s required for fully autonomous vehicles. It would make it easier, or more capable, reduce weird/odd behavior in edge cases, but I don’t think it’s required. I’m not exactly well suited to make that a hill I’m willing to die on, I just don’t think it’s required. They were able to create an autonomous vehicle and drive it across the country in the ‘80s. I think a lot more is possible without an extensive infrastructure. 
 

But really - I want to address the idea of “safe”. Not saying this is you, just a jumping off point on the topic. 

 

often when people are against something they’ll use that word and it’s a problem because it’s just loaded in what it means. There are people that would view any death/injury that can be blamed on an autonomous system as reason the system shouldn’t exist or be used. But that seems foolish to me. In this specific context, lots of people die in automobile accidents. If autonomous vehicles decreased that by 20%, that would be a win. And I think it’s more reasonable to think it would decrease them by significantly more, especially since almost all accidents are (by nature) from someone not following the rules. An autonomous system should follow the rules at a much higher rate. And I would expect that to net out very well against number of issues caused by autonomous vehicles. 
 

I expect the day autonomous vehicles become a real thing, we’ll have loads of people doing this. Pointing at each individual incident with outrage, completely ignoring the bigger picture. I expect that will also cause setbacks in the rollout. 
 

but it could almost eliminate deaths from dui, texting while driving, medical emergencies, exhausted/tired drivers, and stupid stuff like changing the radio station. 
 

but yeah the infrastructure would make it even better. But really I would expect that to make the overall system better for traffic routing and such, not make a cars individual movements better. And I’m not really sure you can offboard the important decision making - I don’t think you can depend on link latency there. I think you need immediate decision making. 

 

With regards to it being required, I think with more limited autonomous driving frameworks you don't actually need all of the surrounding infrastructure. But what I was thinking of was the true longer term vision of fully autonomous driving where people basically do nothing and every car is fully automated. In order to get to that point I think you absolutely have to have smart infrastructure, for multiple reasons. But I agree that it's not required for all types of autonomous driving.

 

Another reason I say required is more about efficiency and the vast amount of data and processing required in many cases. What would be more efficient: having a HD camera in the car, running a ResNet50 CNN on the video it brings in that can tell if the light is green, yellow, red, if it's cloudy, the angle, the light level, if there are any obstructions, etc...or just having the stoplight tell the car that it's red?

 

With regards to security, I completely agree with you when it comes to throwing the baby out with the bathwater. If it can cut down on tons of other potential hazards (some of which you mentioned) then a slight uptick in other hazards due to the occasional corner case malfunction would be acceptable. I was more thinking about it in holistic terms of having such a potential single point of failure for an entire system. If the car is doing all of the processing and something goes completely haywire, then that car has potentially just turned into a 60 mph kinetic weapon that could cause who knows how much damage. Obviously we'd hope there would be some sort of failsafe mechanisms, but it's still a potentially major issue.

 

That would also be something that would drive lawyers up the wall...when someone falls asleep while driving and the car turns into a missile and kills people, the car manufacturer isn't to blame. When that happens due to a computer malfunction, it's a different story. Not saying it's right or wrong, but it's something that has to be taken into account.

 

To the final point, I was going to include something about that in my post, but didn't want to get too into technical weeds and bore everyone to death, but you're right. There are certain systems that will be so latency sensitive that they have to be on the car. Collision avoidance is one. That's when anything above 1ms is basically game over. The computing for autonomous cars will be along a continuum. The most latency sensitive will be on board, then the next level up will be the surrounding infrastructure, then regional for larger scale traffic management, then cloud for long term analysis, etc.

 

[TradeTheBeal!]

 

[mistertim]

18 minutes ago, tshile said:

I also don’t like the airplane analogy. Air is a 3d space that relies on air traffic control to create and maintain lanes on the fly (though they obviously have rules about they do it) to avoid collisions. 
 

 Cars don’t that have. For all intents and purposes that’s a 2d space. And theoretically, in terms of safety, everything you need to know is right around you. Because the system was designed for humans, and it’s unreasonable to expect humans to make safe decisions without having immediate (and salient) aids to assist. That’s why we have the speed limit posted every so many feet even though it didn’t change; or, where appropriate, a sign that warns you that there’s another sign coming up that you need to be aware of but otherwise wouldn’t have enough time to safely process and act (like a stop sign just around the bend, that you can’t currently see.)

 

That said the 3rd party infrastructure could be anything and it’s possible we over exaggerate what it would look like. 
 

It’s entirely possible someone comes up with something you put in the paint used to line the roads that significantly improves the situation. Setting boundaries or signaling some important information (like hey this is now a work zone; speed limit 25mph; not 45 like your gps thinks it is; or in 250 feet it’s one way traffic with flag people controlling it so be prepared for that)

 

Yeah it's certainly not a perfect analogy, but the overall complexity is what still makes it apt IMO. Yes, cars don't have the 3D space to worry about, but they make up for that in the sheer number of extra variables they encounter regularly which planes don't have to worry about. To me the idea of tons of autonomous cars whizzing around without any connection to the surrounding infrastructure is a bit scary.

 

The smart paint idea is very cool, and I think we'll start seeing things like that as computing gets more and more efficient/small and we then get into the realm of nanotech, etc. I actually think we'll eventually see things that are more along the lines of "universal processing substrates" that can more or less be painted on to things and can then process information. That's when you get into true "ambient computing" territory.

[Captain Wiggles]

Quote

"Myself and @limpbizkit are available for any help you may want or need at Twitter. Happy Holidays"

 

[mistertim]

1 minute ago, TradeTheBeal! said:

 

 

How to know your life has taken a turn down a very poor path: You're getting clowned publicly by Fred ****ing Durst.

[CousinsCowgirl84]

1 hour ago, mistertim said:

 

True autonomous driving (like we see in the movies) will never be a reality until the surrounding infrastructure is also as smart as the cars are. Trying to do all processing in the car is a fool's errand and creates single points of failure that are unacceptable in a life or death situation.

 

The car needs to be able to get data from and communicate with the surrounding environment in order to make the correct decisions.

 

Not doing it that way is like pilots flying planes into airports without any air traffic control systems, based purely on on-board radar and sight alone. Not happening.

Even if what you say is true it doesn’t address how Tesla hurt autonomous driving.
 

 

i disagree with what you are saying and every player in the autonomous driving space disagrees with your assessment. Waymo has successfully developed cars that are fully capable of driving themselves. Yesterday I linked to a trucking company that was able to achieve the same thing. Tesla fsd beta works 95 percent of the time, just don’t take an unprotected left. It would make it easier, but it isn’t necessary.

Edited December 21, 2022 by CousinsCowgirl84

[skinsfan4128]

4 minutes ago, CousinsCowgirl84 said:

Even if what you say is true it doesn’t address how Tesla hurt autonomous driving.
 

 

i disagree with what you are saying and every player in the autonomous driving space disagrees with your assessment. Waymo has successfully developed cars that are fully capable of driving themselves. Yesterday I linked to a trucking company that was able to achieve the same thing. Tesla fsd beta works 95 percent of the time, just don’t take an unprotected left. 

Christ, is there anything that assclown can do to make you stop defending him?

 

Bet you're fun at parties. 

 

HTTR!

[CousinsCowgirl84]

3 minutes ago, skinsfan4128 said:

Christ, is there anything that assclown can do to make you stop defending him?

 

Bet you're fun at parties. 

 

HTTR!

Defending who?  I can defend Tesla’s autonomous driving strategy without defending Musks Twitter folly. It is two different things. And the main gist of the post is that several different autonomous driving programs are a little ahead of Tesla in some areas and they didn’t not require smart infrastructure.

Edited December 21, 2022 by CousinsCowgirl84

[mistertim]

11 minutes ago, CousinsCowgirl84 said:

Even if what you say is true it doesn’t address how Tesla hurt autonomous driving.
 

 

i disagree with what you are saying and every player in the autonomous driving space disagrees with your assessment. Waymo has successfully developed cars that are fully capable of driving themselves. Yesterday I linked to a trucking company that was able to achieve the same thing. Tesla fsd beta works 95 percent of the time, just don’t take an unprotected left. 

 

Whoa. Companies who's primary business model and way of making money is doing all autonomous vehicle processing on-board disagree with me? Stop the presses.

 

And there's a huge difference between having cars that are capable of driving themselves in certain contexts and making the leap to fully autonomous systems with tons of autonomous vehicles on the road.

 

You can certainly do the former without smart infrastructure. The latter is where that gets very dicey. If you have tons of cars doing their own thing without communications with the surrounding infrastructure or even each other, that's a potential recipe for disaster.

 

So yeah, we have some cars that can do fully autonomous driving in certain contexts and within certain limitations. But to make the leap from that to the next level will IMO require making the surrounding environment smart as well.

 

5 minutes ago, skinsfan4128 said:

Christ, is there anything that assclown can do to make you stop defending him?

 

Bet you're fun at parties. 

 

HTTR!

 

To her credit, CC84 has been pretty frank about how bad Musk has ****ed up with Twitter.

Edited December 21, 2022 by mistertim

[tshile]

2 minutes ago, mistertim said:

If you have tons of cars doing their own thing without communications with the surrounding infrastructure or even each other, that's a potential recipe for disaster.

I’m envisioning the beltway where a situation occurs that causes all the cars to be stuck in a loop processing local information and they’re just like moving forward 6 inches, stopping, moving backward 6 inches, reprocessing and repeating 

 

it would be glorious 

[CousinsCowgirl84]

5 minutes ago, mistertim said:

 

Whoa. Companies whose primary business model and way of making money is doing all autonomous vehicle processing on-board disagree with me? Stop the presses.

 

Fair enough, still it seems more feasible to develop onboard systems capable of autonomously driving the way a human can than upgrading millions of miles of infrastructure. Even if that capability is 30-100 years away.

 

 

 

Edited December 21, 2022 by CousinsCowgirl84

[mistertim]

Just now, CousinsCowgirl84 said:

 

Fair enough, still it seems more feasible to develop onboard systems capable of autonomously driving the way a human can than upgrades millions of miles of infrastructure.

 

 

Well, let me backtrack a little bit. As I said in my response to @tshile, the processing for autonomous vehicles will have to exist on a continuum, so there's absolutely stuff that will be processed onboard the car and will need to stay there. I'm just saying that in order to make the next big leaps in autonomous cars, IMO the surrounding infrastructure will need to get smart as well in order to support that.

 

And it's certainly easier to develop something onboard, but that doesn't necessarily negate the need for smart infrastructure in order to make big leaps in autonomous vehicle systems. Building out infrastructure is ****ing hard. That's why there aren't tons of companies doing it vs something onboard or SaaS based. But it's still necessary and the longer term payoffs are also huge.

[tshile]

1 minute ago, CousinsCowgirl84 said:

 

Fair enough, still it seems more feasible to develop onboard systems capable of autonomously driving the way a human can than upgrades millions of miles of infrastructure.

 

 

 

Cities would be the first place. And priority would be given on dense population areas in general, as they’re more complicated.
 

Over time it becomes cheaper and people see value in it and it expands. Much like everything else. 
 

Some of it would also be cheap. One of his examples is a red light that signal it’s red to autonomous vehicles (without relying on it looking red). Which by extension means the light could communicate it’s timing cycle and where it is in the cycle, to anything within whatever range. 
 

we already have a form of this technology where emergency vehicles can communicate to intersections such that they interrupt their cycle and force green lights for the direction of travel of the emergency vehicle and everything else to red so it can move through the intersection safer and faster.  And then flip it back once they clear the intersection. 
 

the vehicles themselves will need to utilize a universal system of communicating with others within so many feet just to create a local network of information to work off of. 
 

as I was saying in another post - we (just people in general) may be way off the mark in what we consider infrastructure for this. 
 

the idea of autonomous agents working together is its own subject and something being worked on extensively and there’s a lot of room for creativity and efficiency improvements. I don’t think it’s reasonable to guess we have any clue what it’ll wind up looking like. Rings of localized information simply being shared, where you have many different sizes of rings for different information, may be all we wind up with. The “infrastructure” could wind up being more akin to infrastructure as code, as opposed to physical infrastructure requiring installation and maintenance etc. 

[CousinsCowgirl84]

10 minutes ago, mistertim said:

 

Well, let me backtrack a little bit. As I said in my response to @tshile, the processing for autonomous vehicles will have to exist on a continuum, so there's absolutely stuff that will be processed onboard the car and will need to stay there. I'm just saying that in order to make the next big leaps in autonomous cars, IMO the surrounding infrastructure will need to get smart as well in order to support that.

 

And it's certainly easier to develop something onboard, but that doesn't necessarily negate the need for smart infrastructure in order to make big leaps in autonomous vehicle systems. Building out infrastructure is ****ing hard. That's why there aren't tons of companies doing it vs something onboard or SaaS based. But it's still necessary and the longer term payoffs are also huge.

The extravehicular communication could be on-board the vehicles. You don’t really need it to be centralized, as those hubs would require a huge amount of processing power and represent a single point of failure.

5 minutes ago, tshile said:

Cities would be the first place. And priority would be given on dense population areas in general, as they’re more complicated.
 

Over time it becomes cheaper and people see value in it and it expands. Much like everything else. 
 

Some of it would also be cheap. One of his examples is a red light that signal it’s red to autonomous vehicles (without relying on it looking red). Which by extension means the light could communicate it’s timing cycle and where it is in the cycle, to anything within whatever range. 
 

 

This isn’t that good of an example imo because telling if a stop light is red or green is a solved problem for autonomous vehicles. We shouldn’t waste resources on solved problems. Cross traffic (or anything that moves) is a much harder problem to solve. We need infrastructure to help with that sort of tasks, the ability for my car to know exactly where your car is and where it is going.

 

I assumed that was what @mistertim ws referring to when he talked about infrastructure upgrades, and transportation network to share data between cars.

 

 

Edited December 21, 2022 by CousinsCowgirl84

[tshile]

Dod is heavy into researching autonomous agents working together. 
 

they’re doing it with very smalls drones because they want to send in swarms of drones to fight instead of people, for multiple reasons. Tons of flockbot videos on YouTube you can see what I mean. 
 

but much of that r&d will be useful in designing a fully autonomous vehicle infrastructure. 
 

just for ****s and giggles - a lot of the early work was using the US military as a model. How to control for losing the drone in command (for whatever reason) and what it looks like to navigate that chaos and regain control, and even reevaluate for new goals with regard to the initial goal (and plan), with a bunch of autonomous agents, wound up being a problem the US military had a pretty good grip on (with people in the field of battle) and it informed lots of the early work. 

7 minutes ago, CousinsCowgirl84 said:

This isn’t that good of an example imo because telling if a stop light is red or green is a solved problem for autonomous vehicles. We shouldn’t waste resources on solved problems

computer vision is incredibly hard. And very process intensive. 
 

creating a significantly easier, more reliable and resilient system, that requires less processing to use, is a huge plus. 
 

it’s not a solved problem. Not in a way that matters. 

Edited December 21, 2022 by tshile

[CousinsCowgirl84]

6 minutes ago, tshile said:

Dod is heavy into researching autonomous agents working together. 
 

they’re doing it with very smalls drones because they want to send in swarms of drones to fight instead of people, for multiple reasons. Tons of flockbot videos on YouTube you can see what I mean. 
 

but much of that r&d will be useful in designing a fully autonomous vehicle infrastructure. 
 

just for ****s and giggles - a lot of the early work was using the US military as a model. How to control for losing the drone in command (for whatever reason) and what it looks like to navigate that chaos and regain control, and even reevaluate for new goals with regard to the initial goal (and plan), with a bunch of autonomous agents, wound up being a problem the US military had a pretty good grip on (with people in the field of battle) and it informed lots of the early work. 

computer vision is incredibly hard. And very process intensive. 
 

creating a significantly easier, more reliable and resilient system, that requires less processing to use, is a huge plus. 
 

it’s not a solved problem. Not in a way that matters. 
 

 

 

Proclaiming a networked system easier, reliable and resilient without any evidence is a bold move.

[mistertim]

6 minutes ago, CousinsCowgirl84 said:

The extravehicular communication could be on-board the vehicles. You don’t really need it to br centralized, as those hubs would require a huge amount of processing power and represent a single point of failure.

 

This isn’t that good of an example imo because telling if a stop light is red or green is a solved problem for autonomous vehicles. Cross traffic is a much harder problem to solve. We need infrastructure to help with that sort of tasks, the ability for my car to know exactly where your car is and where it is going.

 

 

 

I think "centralized" is a bit of a misnomer. It would be densely distributed local and regional processing nodes that wouldn't rely on any central "hubs" like cloud does, so those single points of failure wouldn't actually happen.

 

As far as the processing power, it would certainly require a fair amount, but that's actually a case for why it's not feasible to do everything in the car itself. There's only so much processing power you can pack into a vehicle before running into scaling issues.

 

For the traffic lights, I don't think it's really accurate to say it's a "solved problem." Current autonomous vehicles' stoplight detection relies on many things going right: unobstructed line of sight for example. And I'd assume things like bad weather and angle would factor in as well.

 

Because of that, the current traffic light detection is a nice to have thing when it works, but it can't be counted on to always work, which is what you would need in the case of a fully autonomous system.

 

1 minute ago, tshile said:

Dod is heavy into researching autonomous agents working together. 
 

they’re doing it with very smalls drones because they want to send in swarms of drones to fight instead of people, for multiple reasons. Tons of flockbot videos on YouTube you can see what I mean. 
 

but much of that r&d will be useful in designing a fully autonomous vehicle infrastructure. 
 

just for ****s and giggles - a lot of the early work was using the US military as a model. How to control for losing the drone in command (for whatever reason) and what it looks like to navigate that chaos and regain control, with a bunch of autonomous agents, wound up being a problem the US military had a pretty good grip on and informing lots of the early work. 

 

Yeah those swarms of drones are amazing. I was actually just thinking about the possibility of doing something more simple for autonomous vehicles like setting up the cars as agents in a PSO algorithm or something. That would be an even more decentralized method, though you'd still need communication protocols and some connections to the rest of the environment.

 

There's a ton of interesting stuff going on with this now, you're right.