Motoring Discussion > Electric car news | Miscellaneous |
Thread Author: Crankcase | Replies: 6 |
Electric car news - Crankcase |
I know we've had a couple of boring megathreads from me about my own EV experiences; this perhaps could be a general news thread. I appreciate it might be a small audience, but over time... Anyway, today's excitements - Tesla announce ALL their cars from this point on can have fully automatic driving mode, which is clever enough to take you from one side of the States to the other without touching anything. I can only assume they also mean other cars and pedestrians in that. Needless to say, what they ACTUALLY mean is that it will be built in, but not enabled, as pesky regulations are in the way at the minute. www.bbc.co.uk/news/technology-37711489 Or many many other links of your choice. On the rumour mill, Carlos Ghosn (Nissan/Renault) is making a keynote speech in January. Hints are it will be about Nissan introducing automated technology on the Leaf, of some sort. This stuff is beginning to happen, like it or not. Last edited by: Crankcase on Thu 20 Oct 16 at 09:49
|
Electric car news - spamcan61 |
So they're actually saying that future deliveries of current vehicles will have 'Autopilot' turned off - awesome spin! While this is occurring, Teslas with new hardware will temporarily lack certain features currently available on Teslas with first-generation Autopilot hardware, including some standard safety features such as automatic emergency braking, collision warning, lane holding and active cruise control. Google's own data shows that in fully autonomous mode their cars are averaging an accident every 1500 miles (I'll look for the source data on that). This stuff is very very difficult! Last edited by: spamcan61 on Thu 20 Oct 16 at 10:31
|
Electric car news - rtj70 |
Anyone who can programme a computer using a conventional programming language will know you won't be able to implement something that recognises and responds to hazards because you can't really deal with all the fuzziness of the real world. You will need to resort to AI programming, potentially using something like neural networks. You then need to train your neural network and probably tweak it's programming. To train a neural network you need data from the sensors and cameras... so Tesla is now collecting this data to help it develop it's AI components. Why they didn't do this from the start is what bothers me. That accident in America happened because the AI relied more on the camera data and overrode the radar information that said 'watch out'. Last edited by: rtj70 on Thu 20 Oct 16 at 12:06
|
Electric car news - rtj70 |
So to show how hard this is, just think of what you as a driver has to do to avoid hazards. Examples you might need to slow down for or stop are: - Someone crossing the road (adults, children, elderly - take longer to cross) - Animals - Other traffic But even recognising an adult (could be tall/short, fat/thin, black/Asian/caucasian, young/old. male/female) is more complex to emulate with a computer. Hence my 'fuzzy logic' reference and AI. And would you want an automated car to stop if there was a mannequin next to the crossing? If something ran out in front of the car, should it stop if it can to avoid it? What if that causes a problem behind... And should it try to stop if it's a dog but if it's a cat or squirrel then it might just have to carry on. Should it try to swerve? But what if that then causes another danger - to avoid one person in the road you might hit someone else. So do you have a worst case scenario reaction that says if I avoid the child I hit the pensioner and that's what you need to do. But what if it's a pensioner I need to avoid and there's some school kids where I'd need to swerve to? This automated car lark is very complicated and there are many interacting scenarios to deal with. We're not there yet but getting there. |
Electric car news - commerdriver |
>> This automated car lark is very complicated and there are many interacting scenarios to deal >> with. We're not there yet but getting there. >> I cannot ever see it getting past 80 - 90% automated in mixed traffic, i.e. situations where there are pedestrians, animals or vehicles with drivers, even with AI development. And, I suspect, that in many cases the penalty will be lower traffic speeds in many situations as the tendency will be to have settings which are more cautious than many drivers. |
Electric car news - rtj70 |
A good driver is doing a lot in anticipating things that might happen and be ready to stop or whatever. How you programme a car's AI to deal with all this makes it ever more complicated. And yet without it the system cannot be 100% automatic. Examples: - Seeing someone in a car that might suddenly open a door. - A cat/squirrel/whatever moving along a pavement and go under a car and might run across the road - A driver approaching a junction too fast and might not slow down/stop at the junction/give-way line - The wonky hub cap on the car in front that might suddenly be thrown off - Something that's happening way up the road and not immediately in front of the car Remember all of the things a human would and should do needs to be recreated as a conventional programme (so the driving bit which is procedural) or AI (the responding to the environment bit). |
Electric car news - Manatee |
>> A good driver is doing a lot in anticipating things that might happen and be >> ready to stop or whatever. And so is a crap driver, or they would be crashing every 5 minutes. The other thing the yuman brain especially in regard to vision is very good at is "ignoring" information. It would be impossible to see (in the visual, rather then the purely optical sense) everything in the field of view, or to process it continually. Sometimes information falls down the cracks, as when an obstacle/pedestrian/cyclist/motorcyclist is missed because of saccadic suppression. This should be taught to all learners. Most of us eventually learn that looking once is not enough, but don't know why. |