|
Ranter posted:In perfect weather it has existed for years. Even until last year, Google's solution had issues with rain.
|
#
?
Oct 26, 2015 15:13
|
|
|
|
| # ? May 15, 2024 13:08 |
|
|
Subjunctive posted:And required very detailed map information, I believe. I'm changing my vote: Apple's project is doomed.
|
|
#
?
Oct 26, 2015 15:37
|
|
|
Godholio posted:I'm changing my vote: Apple's project is doomed. Perhaps Apple will do the Apple thing and build a car without any automation, then when everyone else has worked out the bugs they leapfrog them to the next level.
|
|
#
?
Oct 26, 2015 18:55
|
|
|
Subjunctive posted:And required very detailed map information, I believe. Yeah, my understanding is that the cars are pre-loaded with detailed 3D scans of every roadway they're going to drive on. I can't imagine that being a practical solution for a final consumer product - even if Google managed to scan all the millions of miles of roads in the US, a few cars would inevitably run into newly built or modified roads before they could update their fancy 3D maps.
|
|
#
?
Oct 27, 2015 01:19
|
|
|
Cockmaster posted:Yeah, my understanding is that the cars are pre-loaded with detailed 3D scans of every roadway they're going to drive on. I can't imagine that being a practical solution for a final consumer product - even if Google managed to scan all the millions of miles of roads in the US, a few cars would inevitably run into newly built or modified roads before they could update their fancy 3D maps. Also even then they're wary of going anywhere with hills or major elevation changes.
|
|
#
?
Oct 27, 2015 01:21
|
|
|
Visual signage and lane markings are the best we can do for humans. Signage and lane markings made to be machine readable from the start seems like a good first step. Then you need all cars on the road to be automated, communicate with each other their course, speed, and intended maneuvers, and be equipped with proximity sensors. Now you have a good, efficient, fast, safe transportation system... until someone goes into traffic without a radio beacon and all other traffic grinds to a halt for safety reasons. I don't think a system that mixes human-operated and machine-operated vehicles will ever be any good.
|
|
#
?
Oct 27, 2015 03:40
|
|
|
pun pundit posted:Visual signage and lane markings are the best we can do for humans. Signage and lane markings made to be machine readable from the start seems like a good first step. Then you need all cars on the road to be automated, communicate with each other their course, speed, and intended maneuvers, and be equipped with proximity sensors. Now you have a good, efficient, fast, safe transportation system... until someone goes into traffic without a radio beacon and all other traffic grinds to a halt for safety reasons. Machines can already read the same signage and lane markings humans do, and not just standard stuff, but things like house numbers. Existing self-driving or assisted-driving cars don't rely on other cars being beaconed or communicating either, they use RADAR and SONAR. The system exists today and the terrible parts are basically the human drivers. Google's Waze acquisition has been really valuable for getting updated GPS-surveyed map data, with proper signage read in and everything. There's an interchange near me (826/836 for any Miami readers) that changes shape every few months of construction, and Waze has the changes within hours.
|
|
#
?
Oct 27, 2015 05:06
|
|
|
If they could do that in real time and under adverse conditions, self-driving cars wouldn't have the terrible results they do in any conditions but flat terrain and pristine weather.
|
|
#
?
Oct 27, 2015 05:54
|
|
|
"Hours" is nowhere near good enough. "Hours" kills hundreds of people per year, easily.
|
|
#
?
Oct 27, 2015 06:27
|
|
|
Godholio posted:"Hours" is nowhere near good enough. "Hours" kills hundreds of people per year, easily. Human drivers kill way more than that per year
|
|
#
?
Oct 27, 2015 06:46
|
|
|
Godholio posted:"Hours" is nowhere near good enough. "Hours" kills hundreds of people per year, easily.
|
|
#
?
Oct 27, 2015 07:00
|
|
|
Humans have problems when roads change, too. For example, when a traffic signal is added to location that previously didn’t have one, locals run red lights for weeks. “Hours” may still be unacceptable, but if self‐driving cars become widespread, I guarantee it will no longer take hours. By the time the cones are out of the roadway, the update will have already been pushed to every car in the area. roomforthetuna posted:It's not like self-driving cars are just going to jump off bridges and poo poo if the road isn't identical to what they have mapped. Humans just might, though. Platystemon fucked around with this message at 08:25 on Oct 27, 2015 |
|
#
?
Oct 27, 2015 08:20
|
|
|
There is a massive legal and moral difference between a human acting stupidly and a machine claiming to do a human's job being temporarily incapable. The car company solves this by saying you should hold your hands on the steering wheel at all times and be ready to take over on a second's notice and be used for highways only. I, like so many others, will give zero shits about that after a week's use and will be happily texting, eating, jerking off or whatever. What used to simple accidents are now suddenly complicated accidents. Since I have to pay attention anyway, even hold the steering wheel, this entire system can be replaced with ancient car technology; an adjustable steering wheel which makes it easy to knee-steer or a center armrest which means I can steer with my thumb while my arm is completely relaxed. Add some nice tunes or podcasts, I'm good for 12 hours on the road. If I can't legally, technically or safely curl up in the backseat for a snooze or scoot over to the passenger seat to piss in a bottle, it's not an autopilot.
|
|
#
?
Oct 27, 2015 09:49
|
|
|
After reading a bit about the new laws about electric cars having to make noise in order to be safer for pedestrians, I found this: https://www.youtube.com/watch?v=ODmsJqebmDo I like it so far.
|
|
#
?
Oct 28, 2015 10:06
|
|
|
Cockmaster posted:Yeah, my understanding is that the cars are pre-loaded with detailed 3D scans of every roadway they're going to drive on. I can't imagine that being a practical solution for a final consumer product - even if Google managed to scan all the millions of miles of roads in the US, a few cars would inevitably run into newly built or modified roads before they could update their fancy 3D maps. Actually they will use onboard LTE connections to download up-to-date HD maps, and upload data constantly to crowdsource those same maps. At least that's what Tesla and HERE are both planning on doing.
|
|
#
?
Oct 28, 2015 13:09
|
|
|
Can these systems use their machine vision to tell what road they're on and use that for navigation? Or cross reference what they see with what they're supposed to see, and if it doesn't match, to find a match? I ask because right now Waze can't tell the difference between driving on a raised highway and driving on a surface street that runs underneath it, and GPS isn't lane-accurate enough to differentiate between parallel adjacent roads going the same direction.
|
|
#
?
Oct 28, 2015 13:31
|
|
|
Linedance posted:Can these systems use their machine vision to tell what road they're on and use that for navigation? Or cross reference what they see with what they're supposed to see, and if it doesn't match, to find a match? Yeah, that's how they work, here's a four year old article on it: http://spectrum.ieee.org/automaton/robotics/artificial-intelligence/how-google-self-driving-car-works quote:Two things seem particularly interesting about Google's approach. First, it relies on very detailed maps of the roads and terrain, something that Urmson said is essential to determine accurately where the car is. Using GPS-based techniques alone, he said, the location could be off by several meters. It seems that they need prepared data to drive fast and fall back to a extremely slow and cautious mode if they don't have the map, or if the road has changed extensively: http://www.technologyreview.com/news/530276/hidden-obstacles-for-googles-self-driving-cars/ quote:Google often leaves the impression that, as a Google executive once wrote, the cars can “drive anywhere a car can legally drive.” However, that’s true only if intricate preparations have been made beforehand, with the car’s exact route, including driveways, extensively mapped. Data from multiple passes by a special sensor vehicle must later be pored over, meter by meter, by both computers and humans. It’s vastly more effort than what’s needed for Google Maps.
|
|
#
?
Oct 28, 2015 14:32
|
|
|
Grim Up North posted:If a new stop light appeared overnight, for example, the car wouldn’t know to obey it. welp.
|
|
#
?
Oct 28, 2015 14:51
|
|
|
Linedance posted:Can these systems use their machine vision to tell what road they're on and use that for navigation? Or cross reference what they see with what they're supposed to see, and if it doesn't match, to find a match? Yes, there's a lot of sensor fusion that goes on. Ford's cars use a combo of LIDAR, GPS, and on-board gyroscopes: http://arstechnica.com/cars/2015/08/face-to-face-with-fords-self-driving-fusion-hybrid-research-vehicles/
|
|
#
?
Oct 28, 2015 14:51
|
|
|
MrOnBicycle posted:After reading a bit about the new laws about electric cars having to make noise in order to be safer for pedestrians, I found this: So they are doing Ford's "fake engine soundinator" thing, except pumping it to the outside instead of the inside? I'm not a fan, I like the coming of completely (well, not really completely) silent cars.
|
|
#
?
Oct 28, 2015 16:37
|
|
|
The Locator posted:So they are doing Ford's "fake engine soundinator" thing, except pumping it to the outside instead of the inside? I'm not a fan, I like the coming of completely (well, not really completely) silent cars. Haven't heard Fords take on it. As long as it isn't a stupid beep that can't be controlled, I'm happy. Besides, "silent" cars won't happen it seems. At least not in city traffic and those speeds. Law is coming in 2018 for the US. Having sound going into the cabin via speakers is the dumbest thing ever though.
|
|
#
?
Oct 28, 2015 17:07
|
|
|
The Locator posted:I like the coming of completely (well, not really completely) silent cars. In our litigious public society full of generally unaware fools? Good luck, you're overruled.
|
|
#
?
Oct 28, 2015 17:52
|
|
|
http://youtu.be/sZJjTEmXaf8 RE: silent cars
|
|
#
?
Oct 28, 2015 18:27
|
|
|
Chris Urmson: How a driverless car sees the road | TED
|
|
#
?
Oct 28, 2015 18:43
|
|
|
How does Elon justify customers paying $2,500 for enabling a feature in software? Referring to Autopilot. $2,500 add-on during purchase, or $3,000 to enable after delivery.
|
|
#
?
Oct 28, 2015 19:40
|
|
|
Same way anyone justifies selling software, I imagine.
|
|
#
?
Oct 28, 2015 19:51
|
|
|
Ranter posted:How does Elon justify customers paying $2,500 for enabling a feature in software? Referring to Autopilot. $2,500 add-on during purchase, or $3,000 to enable after delivery. Have you even heard of Apple?
|
|
#
?
Oct 28, 2015 22:15
|
|
|
Ranter posted:How does Elon justify customers paying $2,500 for enabling a feature in software? Referring to Autopilot. $2,500 add-on during purchase, or $3,000 to enable after delivery. The same reason you can't just make a satellite dish yourself and capture all the signals that are buzzing around your head already. They are already there but you gotta pay to unlock them.
|
|
#
?
Oct 28, 2015 22:17
|
|
|
Ranter posted:How does Elon justify customers paying $2,500 for enabling a feature in software? Referring to Autopilot. $2,500 add-on during purchase, or $3,000 to enable after delivery. What you charge a customer for literally fuckin anything on earth does not have to be "just" in any sense of the word.
|
|
#
?
Oct 28, 2015 23:33
|
|
|
Ranter posted:How does Elon justify customers paying $2,500 for enabling a feature in software? Referring to Autopilot. $2,500 add-on during purchase, or $3,000 to enable after delivery. It's not some foreign concept business model. At least they aren't forcing you to spend $10k on an options package you want nothing else from in order to get it, which is a major annoyance with every other manufacturer. What would be interesting to see is if they pursue a subscription model like software companies have done, so that instead of an upfront fee, you pay as long as you're using it and can cancel your Autopilot subscription any time. And when you trade in for a new model, your software subscription follows you.
|
|
#
?
Oct 28, 2015 23:45
|
|
|
It seems a bit steep from the perspective of a poor schlub who sees that as about a year worth of Leaf payments, and that's still a week's salary for someone making low six figures, but I can't imagine this is a hugely steep price for the majority of Tesla owners, especially if you can buy it as an option when you first get the car. Now weird that that last part has a $500 fee but they can charge it so why not  Not to mention it's a roundabout way of paying for the hardware (nevermind R&D and data transfer costs) that's included standard so that the feature can work in the first place, since it's not just something that can be thrown on aftermarket.
|
|
#
?
Oct 29, 2015 00:01
|
|
|
Boten Anna posted:It seems a bit steep from the perspective of a poor schlub who sees that as about a year worth of Leaf payments, and that's still a week's salary for someone making low six figures, but I can't imagine this is a hugely steep price for the majority of Tesla owners, especially if you can buy it as an option when you first get the car. Yeah I imagine a lot of people are leasing the car so they will look at it as just $20 more a month or whatever it comes out to be
|
|
#
?
Oct 29, 2015 01:01
|
|
|
IIRC from talking to the guy at the store, more than 3/4s of Teslas are purchased rather than leased, which I found surprising at the time. But yeah it's like 2.5% of the purchase price or whatever, I would be pretty surprised if many people were buying without the autopilot package at this point.
|
|
#
?
Oct 29, 2015 01:17
|
|
|
Subjunctive posted:IIRC from talking to the guy at the store, more than 3/4s of Teslas are purchased rather than leased, which I found surprising at the time. But yeah it's like 2.5% of the purchase price or whatever, I would be pretty surprised if many people were buying without the autopilot package at this point. Especially now that it's actually functional. Before, you were paying for a feature that hadn't actually been released yet.
|
|
#
?
Oct 29, 2015 01:37
|
|
|
It was always used for TACC, and the sensors for parking assist/speed limit tracking, I believe.
|
|
#
?
Oct 29, 2015 01:41
|
|
|
Linedance posted:It's not some foreign concept business model. At least they aren't forcing you to spend $10k on an options package you want nothing else from in order to get it, which is a major annoyance with every other manufacturer. Hell, it's less than what Toyota charges for the Advanced Technology Package on the Prius (which is only offered on the top trim level, and is basically a much less advanced version of the stuff in the Tesla autopilot system).
|
|
#
?
Oct 29, 2015 02:18
|
|
|
So when will we see 2016 Volts in California? I see that they were able to be ordered, and it looks like January for them arriving in dealerships. Anyone know anything else?
|
|
#
?
Nov 9, 2015 05:05
|
|
|
Linedance posted:GPS isn't lane-accurate enough to differentiate between parallel adjacent roads going the same direction. GPS can be that accurate (and more!), but your consumer-level GPS in your phone or garmin isn't going to be.
|
|
#
?
Nov 9, 2015 05:25
|
|
|
Frinkahedron posted:GPS can be that accurate (and more!), but your consumer-level GPS in your phone or garmin isn't going to be. Sure, lets just set up D-GPS stations every quarter mile along every road. I'm sure that will solve all of our (or rather AI) problems.
|
|
#
?
Nov 9, 2015 05:42
|
|
|
|
| # ? May 15, 2024 13:08 |
|
|
Brigdh posted:Sure, lets just set up D-GPS stations every quarter mile along every road. I'm sure that will solve all of our (or rather AI) problems. Or subscribe to omnistar, 10cm accuracy or less! Frinkahedron fucked around with this message at 07:47 on Nov 9, 2015 |
|
#
?
Nov 9, 2015 07:43
|
|