4 ms·

It's clear from that wall of text that you're not super familiar with autonomous vehicles. That's fine, no one knows everything. I'll try to clear up a few of t

It's clear from that wall of text that you're not super familiar with autonomous vehicles. That's fine, no one knows everything. I'll try to clear up a few of the more glaring misconceptions. First, the idea that you rarely need to update maps is wrong. The minimum you can really get away with is monthly. Weekly is preferable. Daily is the standard, with serious pressure from operations and government relations support up-to-the-minute updates to handle things like avoiding emergency services routes, route obstructions like downed trees, traffic updates, and unannounced construction zones detected by another vehicle. Secondly, you missed the "data offload" part of what I said. One of the main limitations on operating time for autonomous vehicles isn't charging, it's running out of space to store collected data. This includes data that will be used to update maps, detailed logs of how the vehicle is running and what errors sensors are encountering, as well as the basic sensor outputs and analysis results (e.g. where other vehicles are in space). This is often terabytes after compression and log reduction. Good luck loading that to a USB drive in any reasonable time. Thirdly, good luck making USB devices secure against evil maid attacks. It's a hell of a lot easier and there's a hell of a lot more bandwidth available with mutually authenticated wireless APs. As for taking Waymo to a northern state, they've been testing in Tahoe for years and they used to test in New York. This coming winter they'll be doing heavy testing in both of those as well as Northern Michigan (including UP). Waymo are autonomous. They're not on rails or anything else you've heard. The computers are given a destination and route themselves to it, handling all driving tasks along the way. When remote assistants step in, all they can do is augment the robot with new observations. It still has to do all of the driving in light of that new information.

b112 2y ago
First, the idea that you rarely need to update maps is wrong. No, it isn't. Many people do not use Google Maps, or any maps like technology as they drive about. They rely upon road signs, and signs saying "Maintenance from Dec to March", and so on, along with 'detour' signs. Cars can easily read road signs these days, and that can be expanded. Further, people can put in their own "bypass" route, which could even be augmented "For the next week". There is no requirement to update often, except for quite literally made up requirements, beyond the requirements humans have. A true requirement for full self-driving, is the ability to indicate "STOP!" or "Let me off here", or for example "Take this street to bypass this mess in the future", whether verbal of via a console. Self-driving doesn't mean you don't tell the chauffeur your preferences, outside of it reading detour signs. In fact, it's a requirement to take directions from a human in the vehicle, as to destination, route, emergency stops, and so on. And this highlights my point. Dozens of ways to have maps updated less often, very low impact, but suddenly "We need to connect to the most dangerous thing for any computing device, the Internet". -- In terms of data offload, it's not required. If it was an actual self-driving car, it could drive in areas with no mobile or cell service, and yes many such places exist. It could be owned as a self-driving car, where people live "off the grid" entirely. What would happen then? Would the vehicle cease to function due to a full disk? Crash? That's very, very poor design. And to that end, all of that data is essentially not required for operations, but for debug and improvement. And this highlights how beta-ish this tech is. You don't need daily logs and endless updates on something stable. The core components of auto-driving vehicles should be in perm-maintenance only mode, with on additions or changes, that's where stable code ends up. No changes. -- In terms of evil maid attacks, seriously? The mega edge case, compared to the lunacy of placing a device online? Online is dumb, dumb, dumb. It's unsecure beyond comparison. If you have anything you care about online, you're doing it wrong. -- In terms of testing, yes I'm sure they are testing. Yet they're not testing anywhere near their operations in SF, and the number of Waymos on test drives around Palo Alto dwarfs anything in a snow laden environment. Testing doesn't mean they've managed to make any headway against blizzard conditions, against snow on road, against unmapped roads with snow on them, against the car icing up, on and on and on. Testing != working. -- And in terms of your comment about remotes, I am unsure of the validity of your statement, but regardless, helping it determine landmarks (aka new observations) is indeed my point. SF is the most mapped area on the planet. And yet people still have to step in and provide guidance. Self-driving is a joke at this point, the best is barely 4 as per J3016, and this chart is a best-describe: https://images.synopsys.com/is/image/synopsys/levels-of-driving-automation?qlt=82&wid=1200&ts=1713958060283&$responsive$&fit=constrain&dpr=off https://images.synopsys.com/is/image/synopsys/levels-of-driv... Note the 'geofencing is required' in this specific definition, which is key to how immensely limited Waymo is, and I'd take this further and say Waymo is a '3.5' at best, with 'human is required' being offloaded for emergency intervention/stops if required. (Don't try to tell me Waymo doesn't do emergency remote stops.. it happens. And 'human is required' doesn't only mean 'the one in the car'). Couple this with the fact that it's extremely easy to make something perform correctly some of the time, difficult most of the time, and an immense feat of incredible engineering for all of the time, even when all means a whole slew of 9s, not 100%. True self-driving without all sorts of conditionals is a minimum of 10 years out, most likely closer to 20 years, and I wouldn't be surprised if 30. People have such rose coloured glasses on this front.

b112 2y ago

First, the idea that you rarely need to update maps is wrong. No, it isn't. Many people do not use Google Maps, or any maps like technology as they drive about. They rely upon road signs, and signs saying "Maintenance from Dec to March", and so on, along with 'detour' signs. Cars can easily read road signs these days, and that can be expanded. Further, people can put in their own "bypass" route, which could even be augmented "For the next week". There is no requirement to update often, except for quite literally made up requirements, beyond the requirements humans have. A true requirement for full self-driving, is the ability to indicate "STOP!" or "Let me off here", or for example "Take this street to bypass this mess in the future", whether verbal of via a console. Self-driving doesn't mean you don't tell the chauffeur your preferences, outside of it reading detour signs. In fact, it's a requirement to take directions from a human in the vehicle, as to destination, route, emergency stops, and so on. And this highlights my point. Dozens of ways to have maps updated less often, very low impact, but suddenly "We need to connect to the most dangerous thing for any computing device, the Internet". -- In terms of data offload, it's not required. If it was an actual self-driving car, it could drive in areas with no mobile or cell service, and yes many such places exist. It could be owned as a self-driving car, where people live "off the grid" entirely. What would happen then? Would the vehicle cease to function due to a full disk? Crash? That's very, very poor design. And to that end, all of that data is essentially not required for operations, but for debug and improvement. And this highlights how beta-ish this tech is. You don't need daily logs and endless updates on something stable. The core components of auto-driving vehicles should be in perm-maintenance only mode, with on additions or changes, that's where stable code ends up. No changes. -- In terms of evil maid attacks, seriously? The mega edge case, compared to the lunacy of placing a device online? Online is dumb, dumb, dumb. It's unsecure beyond comparison. If you have anything you care about online, you're doing it wrong. -- In terms of testing, yes I'm sure they are testing. Yet they're not testing anywhere near their operations in SF, and the number of Waymos on test drives around Palo Alto dwarfs anything in a snow laden environment. Testing doesn't mean they've managed to make any headway against blizzard conditions, against snow on road, against unmapped roads with snow on them, against the car icing up, on and on and on. Testing != working. -- And in terms of your comment about remotes, I am unsure of the validity of your statement, but regardless, helping it determine landmarks (aka new observations) is indeed my point. SF is the most mapped area on the planet. And yet people still have to step in and provide guidance. Self-driving is a joke at this point, the best is barely 4 as per J3016, and this chart is a best-describe: https://images.synopsys.com/is/image/synopsys/levels-of-driving-automation?qlt=82&wid=1200&ts=1713958060283&$responsive$&fit=constrain&dpr=off https://images.synopsys.com/is/image/synopsys/levels-of-driv... Note the 'geofencing is required' in this specific definition, which is key to how immensely limited Waymo is, and I'd take this further and say Waymo is a '3.5' at best, with 'human is required' being offloaded for emergency intervention/stops if required. (Don't try to tell me Waymo doesn't do emergency remote stops.. it happens. And 'human is required' doesn't only mean 'the one in the car'). Couple this with the fact that it's extremely easy to make something perform correctly some of the time, difficult most of the time, and an immense feat of incredible engineering for all of the time, even when all means a whole slew of 9s, not 100%. True self-driving without all sorts of conditionals is a minimum of 10 years out, most likely closer to 20 years, and I wouldn't be surprised if 30. People have such rose coloured glasses on this front.