Autonomous Systems

Palum · Aug 22, 2016

My favorite is the 'slow down or move over' law which doesn't actually describe by how much one should slow down or move over if you see flashing lights. So basically, you know, figure it out assholes.

Blakkheim · Aug 22, 2016

Cybsled said:
The biggest problem with driving is most people think they are good drivers, but they aren't. They're the fuckheads that pull into the middle lane of a highway going 40 miles an hour, the dipshit in a SUV who thinks because they are in a SUV they can drive 80 mph in a blizzard with heavy snow accumulation on the highway, the left lane bandit, the assholes that don't understand zipper merge, the people who slow to 50 MPH when they see a cop even though the speed limit is 65 and cops usually don't give a shit unless you are pushing 80, the "offisher i am only buzzed, not drunk", etc. This is why we have tens of thousands of people die or get maimed every year in MVA.

I totally get that some people like driving, but like the earlier analogy with horses, people will still be able to self-drive, but they will be the minority and restricted as to where they can do that.

We should simply be able to give other drivers a rating. Accumulate enough "Asshole" and "Fucking Asshole" ratings and you eventually get your license suspended for a month. Then 6 months. Then two years. etc. Your insurance company will also get to see how many "Assholes" you got during the last six months and will adjust your rates accordingly.

Tuco · Aug 23, 2016

Blakkheim said:
Well put this in your pipe and smoke it.

End-to-End Deep Learning for Self-Driving Cars

Cool paper.

Mono and stereo cams play a big role in autonomy, but outdoor computer vision (CV) is hard. People don't realize how great our eyes are, with all their cones and rods. The human eye is really amazing and our cameras can beat it in some ways, but in handling variety they are no where close. You know how when you're driving and the sun is low on the horizon and you can barely see shit? This is a massive failure mode for on-road CV. Bright winter day with a bunch of leafless trees nearby? All your edge detection algorithms to find cars, people, roads, signs etc are now taking a shit.

People also don't realize how much a role our brain plays into comprehension of the image. We see a bus at 250m away, to us it's obvious that it's a bus and there's no real thought required. But you know how sometimes you see a trailer hauling some weird piece of factory machinery? You see it at a distance and try to figure out wtf it is. You can't perceive it because it's this mangled mess you don't understand. Only when you get close can you start to make out details about the structure. Or you know how you glance left while driving, and think you see a person? You even think you see details about their clothes, face, movement etc. Then you check again and see it's a mailbox/tree/bush/whatever? That's not your eyes playing tricks on you, it's your brain. And it's fucking awesome.

Learning systems like the nvidia one play a big role in trying to build and tune CV, and generally they work with constraints. But each parameter and factor is a variable the learning system has to account for. You keep adding variables and you increase the data set required to account for them, the cost in learning from that data and the cost in processing new data increases exponentially.

This sounds like it's just a matter of throwing my hardware at it to solve, but it really isn't because you start to run into boundaries of learning systems that turn into a big lop of undependable shit when they have a high enough cost for collision and are afraid that at any moment a lady in a electric wheelchair can chase a duck across the road in front of them (This is a famous example google shows off).

ZyyzYzzy · Aug 23, 2016

Tuco said:
Cool paper.

Mono and stereo cams play a big role in autonomy, but outdoor computer vision (CV) is hard. People don't realize how great our eyes are, with all their cones and rods. The human eye is really amazing and our cameras can beat it in some ways, but in handling variety they are no where close.

You know how when you're driving and the sun is low on the horizon and you can barely see shit? This is a massive failure mode for on-road CV. Bright winter day with a bunch of leaveless trees nearby? All your edge detection algorithms to find cars, people, roads, signs etc are now taking a shit.

People don't even realize how amazing even the eyesight of a god damn fly is. Literally trying to parse the optic neuronal processong of drosophila was important for guided missles. Tracking something simple in a complex environemt is something still relatively new.

AVCPL · Aug 23, 2016

Delphi, Mobileye to offer driverless car system in 2019
Each company is bringing different technologies to the table to create the driverless car system. Mobileye specializes in real-time mapping, while Delphi specializes in software and sensors, including cameras, LiDAR and radar. The two companies say automakers will be able to buy the system and install it on a range of vehicle platforms, from small cars to SUVs and crossovers and could offer them to the public as soon as late 2019 or 2020.

Picasso3 · Aug 23, 2016

Leafless.

Eyesight on subaru kicks off in heavy rain which is of course no problem and good policy. 99% autonomy is fine imo, especially on predictable times like dusk and inclement weather. The tesla guy flying into the tractor trailer is brutal, but I bet they'll have most of those scenarios ironed out after a few hundred deaths.

Tuco · Aug 23, 2016

AVCPL said:
Delphi, Mobileye to offer driverless car system in 2019
Each company is bringing different technologies to the table to create the driverless car system. Mobileye specializes in real-time mapping, while Delphi specializes in software and sensors, including cameras, LiDAR and radar. The two companies say automakers will be able to buy the system and install it on a range of vehicle platforms, from small cars to SUVs and crossovers and could offer them to the public as soon as late 2019 or 2020.

I got to see the delphi system recently. It's very cool. I'm really looking forward to seeing it and all the other systems integrated.

I'm betting that the driving behavior of autonomous cars will be very noticeable. My guess is that it will be very cautious.

Tripamang · Aug 23, 2016

Cool quoting works from other threads~ From the political thread

mkopec said:
LOL, youre living in la-la land bro. A decade? Maybe 25-50 yrs I would give you but a decade? Not gonna happen.

I don't think I am, mainly because the most expensive part of transportation is the human element. If you can eliminate humans AND keep your vehicles running 24/7 because they no longer have a driver who needs rest, well you're in an excellent position to disrupt the market. The cost savings to adopt these things is going to be so immense, that no one can afford not to adopt them as quickly as possible. If somebody can ship something from A->B at 1/10th the cost, you simply can't compete against that. The only deciding factor on how long it takes, is how fast they can be built.

Tuco said:
I disagree. If you want to continue, I'll respond here:
Autonomous Vehicles

I'm not saying these things are going to be perfect, but having a long haul truck that goes between cities autonomously and driven locally by drivers would be perfectly viable in 5 years, and anyone who isn't using it within 10 simply won't be able to compete.

Tuco · Aug 23, 2016

My argument is this:
1. It's unlikely that Peterbilt, Daimler, Freightliner, Kenworth etc are 10 years away from selling inter-city autonomy commercial vehicles. Perfectly viable in 5 years is fiction.
2. Even if #1 was false, it's unlikely the first few versions over 5-10 years would cost less to operate than the meatbag vehicles. Truck drivers are not expensive, will become less expensive as autonomy threatens them (after much anguish) and will still be necessary as autonomous vehicles enter the scene. 1/10th the cost is ludicrous.
3. Even if #2 was false, FedEx, Schneider etc wouldn't just start buying them and replacing their fleet asap. They'd try them out over a period of several years, and would phase them in over time as their current fleet expired.
4. Even if #3 was false, it will take several years to build up the infrastructure required to support autonomous fleets.
5. That infrastructure wouldn't just spring up globally/nationally, they would do a trial run over several years in a specific area. This trial run would have a very minor impact on the number of truck drivers nationally.
6. The first autonomous vehicles won't be driverless. They will be driver-assistive vehicles. Shipping companies should be very excited about moving from driver-assisted to driverless, but it won't happen quickly.

So when you see articles like this:
Peterbilt shows off autonomous truck - Truck News

And see that someone has made a very cool truck, don't think we're right around the corner from autonomy.

khorum · Aug 23, 2016

Tuco said:
My argument is this:
1. It's unlikely that Peterbilt, Daimler, Freightliner, Kenworth etc are 10 years away from selling inter-city autonomy commercial vehicles. Perfectly viable in 5 years is fiction.
2. Even if #1 was false, it's unlikely the first few versions over several years would likely cost more to buy and operate than the meatbag vehicles. Truck drivers are not expensive, will become less expensive as autonomy threatens them (after much anguish) and will still be necessary as autonomous vehicles enter the scene. 1/10th the cost is ludicrous.
3. Even if #2 was false, FedEx, Schneider etc wouldn't just start buying them and replacing their fleet asap. They'd try them out over a period of several years.
4. Even if #3 was false, it will take several years to build up the infrastructure required to support autonomous fleets.
5. That infrastructure wouldn't just spring up globally/nationally, they would do a trial run over several years in a specific area. This trial run would have a very minor impact on the number of truck drivers nationally.

So when you see articles like this:
Peterbilt shows off autonomous truck - Truck News

And see that someone has made a very cool truck, don't think we're right around the corner from autonomy.

Tuco fuck all that shit.

Let's get rich cuz I just got the notion of where autonomous vehicles would break into the consumer market: Let's all kickstarter the rerolled ROBO-WINNEBAGO!

You can just cruise the country nonstop until you need to charge its batteries every thousand miles or so. You can cook, watch movies, fuck your wife, play those rang-rang MMOs u love so much, do all your homey things while your Winnebago drives you around the country evading collections agents.

By the time it's ready to hit the market, all of these pasty NEETs will be ready to retire and take out massive high interest loans for our self-driving Poopsock-Mobile. We could even make windows optional but I doubt anyone would buy them.

Preorder now [email protected]

Abefroman · Aug 23, 2016

Tripamang said:
Cool quoting works from other threads~ From the political thread

I don't think I am, mainly because the most expensive part of transportation is the human element. If you can eliminate humans AND keep your vehicles running 24/7 because they no longer have a driver who needs rest, well you're in an excellent position to disrupt the market. The cost savings to adopt these things is going to be so immense, that no one can afford not to adopt them as quickly as possible. If somebody can ship something from A->B at 1/10th the cost, you simply can't compete against that. The only deciding factor on how long it takes, is how fast they can be built.

I'm not saying these things are going to be perfect, but having a long haul truck that goes between cities autonomously and driven locally by drivers would be perfectly viable in 5 years, and anyone who isn't using it within 10 simply won't be able to compete.

No way. The amount of testing needed and the approval process alone will take longer then that. The technology isn't even there yet to even start the other two. You also can't eliminate the drivers for several reasons which eliminates the cost savings. You still need drivers for the city, spotters, someone to make the delivery and deal with the warehouses. Also an entire security force to protect the trucks and fix them if they break down fast.

Their already is a city to city mostly autonomous system in place in the Railroad and many don't use it because truck drivers are very valuable for improvising which happens a mother fucking ton in distribution.

Tuco · Aug 23, 2016

khorum said:
Tuco fuck all that shit.

Let's get rich cuz I just got the notion of where autonomous vehicles would break into the consumer market: Let's all kickstarter the rerolled ROBO-WINNEBAGO!

You can just cruise the country nonstop until you need to charge its batteries every thousand miles or so. You can cook, watch movies, fuck your wife, play those rang-rang MMOs u love so much, do all your homey things while your Winnebago drives you around the country evading collections agents.

By the time it's ready to hit the market, all of these pasty NEETs will be ready to retire and take out massive high interest loans for our self-driving Poopsock-Mobile. We could even make windows optional but I doubt anyone would buy them.

Preorder now [email protected]

Reminds me of what Artamis drove in Ready Player One.

mkopec · Aug 23, 2016

Tripamang said:
I don't think I am, mainly because the most expensive part of transportation is the human element. If you can eliminate humans AND keep your vehicles running 24/7 because they no longer have a driver who needs rest, well you're in an excellent position to disrupt the market. The cost savings to adopt these things is going to be so immense, that no one can afford not to adopt them as quickly as possible. If somebody can ship something from A->B at 1/10th the cost, you simply can't compete against that. The only deciding factor on how long it takes, is how fast they can be built.

Im not disputing that it will not happen. You bet your ass it will. Eventually all meatbag vehicles will be extinct. Its just your time frame of a decade is laughable, thats all.

For one think of the implication of just one of those trucks in a beta stage hitting a school bus, or even a mini van with mommy and 2 kids or some other schlep. That incident alone is enough to push the whole program back a decade or two.

khorum · Aug 23, 2016

It wasn't self-driven tho was it?

If you make it cheap enough to own and maintain, all you really need to do is make it more affordable than rent and utilities. We can even bake in the costs of free national wifi as a loss-leader.

Shit if you can make it cheap enough the whole thing could be a loss leader---we'd just make money off the tax credits for being a plug-in and maybe set up roadside pitstops where we swap out the ROBOBAGO's battery packs, drain their sewage tanks and re-up their hotpocket supplies for a reasonable fee.

fanaskin · Aug 23, 2016

Tuco said:
Cool paper.

Mono and stereo cams play a big role in autonomy, but outdoor computer vision (CV) is hard. People don't realize how great our eyes are, with all their cones and rods. The human eye is really amazing and our cameras can beat it in some ways, but in handling variety they are no where close. You know how when you're driving and the sun is low on the horizon and you can barely see shit? This is a massive failure mode for on-road CV. Bright winter day with a bunch of leafless trees nearby? All your edge detection algorithms to find cars, people, roads, signs etc are now taking a shit.

People also don't realize how much a role our brain plays into comprehension of the image. We see a bus at 250m away, to us it's obvious that it's a bus and there's no real thought required. But you know how sometimes you see a trailer hauling some weird piece of factory machinery? You see it at a distance and try to figure out wtf it is. You can't perceive it because it's this mangled mess you don't understand. Only when you get close can you start to make out details about the structure. Or you know how you glance left while driving, and think you see a person? You even think you see details about their clothes, face, movement etc. Then you check again and see it's a mailbox/tree/bush/whatever? That's not your eyes playing tricks on you, it's your brain. And it's fucking awesome.

Learning systems like the nvidia one play a big role in trying to build and tune CV, and generally they work with constraints. But each parameter and factor is a variable the learning system has to account for. You keep adding variables and you increase the data set required to account for them, the cost in learning from that data and the cost in processing new data increases exponentially.

This sounds like it's just a matter of throwing my hardware at it to solve, but it really isn't because you start to run into boundaries of learning systems that turn into a big lop of undependable shit when they have a high enough cost for collision and are afraid that at any moment a lady in a electric wheelchair can chase a duck across the road in front of them (This is a famous example google shows off).

Why don't they just put transponders into everything and make it part of getting your emissions test? done.

Tuco · Aug 23, 2016

fanaskin said:
Why don't they just put transponders into everything and make it part of getting your emissions test? done.

What?

Kiroy · Aug 23, 2016

Tuco said:
My argument is this:
1. It's unlikely that Peterbilt, Daimler, Freightliner, Kenworth etc are 10 years away from selling inter-city autonomy commercial vehicles. Perfectly viable in 5 years is fiction.
2. Even if #1 was false, it's unlikely the first few versions over 5-10 years would cost less to operate than the meatbag vehicles. Truck drivers are not expensive, will become less expensive as autonomy threatens them (after much anguish) and will still be necessary as autonomous vehicles enter the scene. 1/10th the cost is ludicrous.
3. Even if #2 was false, FedEx, Schneider etc wouldn't just start buying them and replacing their fleet asap. They'd try them out over a period of several years, and would phase them in over time as their current fleet expired.
4. Even if #3 was false, it will take several years to build up the infrastructure required to support autonomous fleets.
5. That infrastructure wouldn't just spring up globally/nationally, they would do a trial run over several years in a specific area. This trial run would have a very minor impact on the number of truck drivers nationally.
6. The first autonomous vehicles won't be driverless. They will be driver-assistive vehicles. Shipping companies should be very excited about moving from driver-assisted to driverless, but it won't happen quickly.

So when you see articles like this:
Peterbilt shows off autonomous truck - Truck News

And see that someone has made a very cool truck, don't think we're right around the corner from autonomy.

Pretty much completely agree - in a decade driver-assisted vehicles will probably start to be pretty common. I think we'll need another decade or two on top of that before driverless start to become common, and probably a few more decades until it actually becomes a safe network of driverless only vehicles, since non-driverless vehicles will be the main error in the system.

It will likely start with freeways becoming driverless only or some such thing, but that's gotta be half a century away just due to the major cultural shift it causes.

mkopec · Aug 23, 2016

Kiroy said:
Pretty much completely agree - in a decade driver-assisted vehicles will probably start to be pretty common. I think we'll need another decade or two on top of that before driverless start to become common, and probably a few more decades until it actually becomes a safe network of driverless only vehicles, since non-driverless vehicles will be the main error in the system.

It will likely start with freeways becoming driverless only or some such thing, but that's gotta be half a century away just due to the major cultural shift it causes.

Agree with this. Too bad we wont be around to see it, or old ass men wetting our diapers by then. Would be wondrous to see though.

Borzak · Aug 23, 2016

Autonomous trucking may work eventually in true point A to point B at a wharehouse or something. Lots of stuff I've done in my life and seen that no computer operated truck is going to be able handle now and not for a while. Everything we order at the shop we have to tell the guy where to park to unload so that we can get X equipment and be out of the way for Y coming up that he can't know about. Happens millions of times a day on jobsites. I won't know exactly where till he drives up, so it's not like I can designate "park here" ahead of time via a computer. I also don't know how you are going to ask a computer "Can you squeeze in here and parly jack knife so we can fit in some shit around you while we unload?" and a 1,000 other things I've asked a delivery driver and that's not counting logging trucks having to play in the mud and such.

AVCPL · Aug 23, 2016

I want to see a fully autonomous tesla in a NASCAR race, programmed to a highly aggressive driving mode.

Autonomous Systems

Would you ever own an autonomous vehicle?

Hell yeah Bring on our robotic overlords!

Fuck you! I'll keep my Indepenence

what Suineg set it to

Karazhan Raider

I got Tuco'd!

RIP USA

N00b

Silver Baronet of the Realm

I got Tuco'd!

Naxxramas 1.0 Raider

I got Tuco'd!

Murder Apologist

Naxxramas 1.0 Raider

I got Tuco'd!

Murder Apologist

Well known agitator

I got Tuco'd!

Marine Biologist

Bronze Baron of the Realm

N00b