|
https://forums.somethingawful.com/showthread.php?threadid=3806088
|
#
?
Apr 21, 2017 00:03
|
|
|
|
| # ? Apr 23, 2024 10:59 |
|
|
BRAKE FOR MOOSE posted:Google Maps is really awful in rural areas with unmaintained winter roads, as just one example. In central VT, NH, ME it will regularly tell you to take roads that will leave you stranded. I see this as entirely solvable, but it unquestionably is a problem that requires a lot of resources. even in the future where self driving cars are feasible and widespread i dont think they'll be as common in rural areas for numerous reasons, not including the increased need to take vehicles off road ironically tho one of the most common contemporary deployments of autonomous vehicles navigating via gps are large farm machines. robot combines, seeders, fertiliziers etc. navigating open areas via gps waypoints
|
|
#
?
Apr 21, 2017 00:42
|
|
|
Buddy I don't really get your thing here and I'm a person who doesn't commute 6 miles to work without waze in front of me. It's great, but not perfect. The point here is that because its not perfect automated cars can't rely on it and if they can't rely on it they need to be able to read street signs including temporary work zone ones on plywood in weird places and solving that problem is the more difficult thing. It's trivial to physically enter data into a database but it's not trivial to get the entire country to do it reliably. Once we have millions of of data-using and/or data-collecting autonomous cars we'll probably have fantastic up-to-date map databases but we're simply not going to be able to rely on those databases to get us there.
|
|
#
?
Apr 21, 2017 01:40
|
|
|
boner confessor posted:did you? you just complained that your gps was wrong a couple of times. i dont see you saying that they can't or won't improve 1) LTE data not available everywhere, infeasible to provide for the coasts: JawnV6 posted:Yeah who even wants to drive anywhere without a solid LTE signal? Nobody I care about, that's for sure. JawnV6 posted:It's also worth noting there are more failure modes than "outdated db" or "highway melted," google maps thought my hotel in China was in the water because they used the national maps. "The corporation wants to play nice with a government uncommitted to high-fidelity data." boner confessor posted:i'm sorry my description of potential future applications of technology differs from your own imagination? im not sure what your problem here is - vehicles ship today with built in gps navigation, gps navigation is also possible with a range of consumer devices. it's not my problem if that irritates you boner confessor posted:and you dont think thousands of vehicles scanning the immediate area and providing adaptive feedback wont work? this is somehow too outlandish in a scenario where we have cars driven by AI? especially considering that an extremely small fraction of that data will actually need to be updated on a daily or even hourly basis, given that speed limits rarely actually change? boner confessor posted:as soon as i ask the question "you've asserted it won't work, but why wouldn't it work" people seem to not want to continue the derail, i wonder why that is? check this out: JawnV6 posted:Having two inputs doesn't reduce the chance for error, it generally doubles it. If an analogy helps: Twin-engine planes that require a skilled pilot to use when one blows out are more dangerous because you've doubled the chance of engine failure, not safer because there's a backup. "Bb..but I added an engine! how could that make it WORSE??"
|
|
#
?
Apr 21, 2017 01:45
|
|
|
asdf32 posted:The point here is that because its not perfect automated cars can't rely on it and if they can't rely on it they need to be able to read street signs including temporary work zone ones on plywood in weird places and solving that problem is the more difficult thing. i never said they wouldn't be able to read street signs  like i said to the other guy the only reason you would think i'm saying "A instead of B" instead of "A and B" is because you can't read or you're bored and looking for an argument. and a centralized traffic alert in a digital format is going to be a better and more reliable option than a spray painted plywood signasdf32 posted:It's trivial to physically enter data into a database but it's not trivial to get the entire country to do it reliably. Once we have millions of of data-using and/or data-collecting autonomous cars we'll probably have fantastic up-to-date map databases but we're simply not going to be able to rely on those databases to get us there. i don't see why not? everyone is repeatedly saying it can't happen but nobody seems to be able to say why. you don't need the entire country to fill out a database, you only need a few central traffic alert systems and like everything else i'm talking about this is technology that already exists and is in use. like not a single thing i'm talking about is any more future technology than self driving cars themselves, all of this - from gps navigation to central road databases to real time traffic and routing information - every single bit of this exists and is use today, right now JawnV6 posted:What a horrible butchering of an attempt to read my post. that's great buddy, i dont know what's annoying you so much but it's not anything that has to do with what i've posted about using gps to navigate so good luck with that
|
|
#
?
Apr 21, 2017 02:24
|
|
|
Because society isnt motivated to maintain a 4 9's grade nationwide map/construction/speed database if there are no cars to consume it and it doesn't solve workzones with hand waving traffic cops anyways so for the 43rd time automated cars have to solve the difficult visual interpretation problems no matter what which makes your database mostly irrelevant.
|
|
#
?
Apr 21, 2017 04:06
|
|
|
JawnV6 posted:Vehicles today don't ship with a radio protocol that can broadcast their assumed position and close objects to other vehicles or a centralized network to synthesize into a coherent whole. You've explicitly relied on "well, what if every car knew what every other car could see!" to duck an imperfect road database and it's lovely sci-fi without any credible implementation. The fact that you see this massive cross-manufacturer protocol development as some obvious thing on par with a GPS device that shows a picture to a human is appalling. You're continually bouncing back and forth between claiming you're describing extant technology and pointing to GPS nav units, then claiming any shortcoming can be handled by 1) advanced AI reading local terrain and signage supported by 2) thousands of vehicles blaring details ranging from "this exit is out" to precise positioning. I don't believe in either of those systems. plus, his vehicle communications network introduces the byzantine generals problem. sure, you've got a large number of vehicles scanning the road, but what about the ones that are malfunctioning or returning malicious results? how can your car even make sure the "vehicles" that are scanning the road around it actually exist and aren't spoofed by some malicious entity?
|
|
#
?
Apr 21, 2017 08:25
|
|
|
boner confessor posted:redundancy. it's also easier to transmit or calculate "go to node 29025 along line 56720 no faster than 40mph, stop, turn left, go to node 570014 along line 34890 no faster than 60mph, stop, turn right" etc. so on because you already have to have this solution in place for a self driving car to figure out how to get to a destination in the first place. like cars aren't going to get where they're going just from sensors alone. how would they know which roads to take? you must supply a route to a self driving car, and it's trivial to supply road metadata along with that route, so... The proper way to implement redundancy in an automated car is to have redundant sensors. "I can't read the signs, so I'll just drive blind based on GPS data and stored metadata" isn't redundancy, it's stupidity. It's not trivial to supply road metadata along that route. That's the problem. It's certainly possible, but it's not necessary or even really useful, so why should the company invest in doing it? I visually watch out for signage, obviously. Even if I know the route, I'm still looking at the road! Think of it another way - when you drive to someplace you don't know the location of, do you spend the entire drive staring at your GPS screen with occasional quick glances at the road, or do you spend the entire drive staring at the road with an occasional glance at your GPS?
|
|
#
?
Apr 21, 2017 15:20
|
|
|
Condiv posted:plus, his vehicle communications network introduces the byzantine generals problem. sure, you've got a large number of vehicles scanning the road, but what about the ones that are malfunctioning or returning malicious results? how can your car even make sure the "vehicles" that are scanning the road around it actually exist and aren't spoofed by some malicious entity? you can validate the results through some central service. iirc tesla currently does this Main Paineframe posted:The proper way to implement redundancy in an automated car is to have redundant sensors. "I can't read the signs, so I'll just drive blind based on GPS data and stored metadata" isn't redundancy, it's stupidity. It's not trivial to supply road metadata along that route. That's the problem. It's certainly possible, but it's not necessary or even really useful, so why should the company invest in doing it? you say it's not trivial to do, but i can buy a device for $50 that does it. there are apps that do it for free. you keep ignoring that the technology you say can't exist or doesn't work currently exists and people use it. it's odd that you don't have any response to this fact
|
|
#
?
Apr 21, 2017 16:13
|
|
|
|
| # ? Apr 23, 2024 10:59 |
|
|
boner confessor posted:you say it's not trivial to do, but i can buy a device for $50 that does it. there are apps that do it for free. you keep ignoring that the technology you say can't exist or doesn't work currently exists and people use it. it's odd that you don't have any response to this fact Can't be all that trivial if you need a $50 dedicated device just to read the metadata, which is much simpler than populating that database and updating it. Yes, I know, your idea for populating and updating is to have the cars do that automatically...but giving them the ability to do that isn't a trivial undertaking either. That doesn't mean it's difficult or impossible, it means that it's something that takes a meaningful investment of time and resources to do, which brings into play the question of whether the payoff for that feature is worth the investment. And since merely having the capability to implement that feature is enough to render that feature entirely obsolete and unnecessary, it doesn't come out of the cost/benefit analysis looking very good!
|
|
#
?
Apr 21, 2017 17:06
|
|