 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
http://www-personal.acfr.usyd.edu.au/spns/motion/CornelSkynet.pdf
Lots of technical detail about how the sensor and control systems work
in an autonomous car. Pay attention - this is the future!
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Le 09/01/2013 11:56, scott nous fit lire :
> http://www-personal.acfr.usyd.edu.au/spns/motion/CornelSkynet.pdf
>
> Lots of technical detail about how the sensor and control systems work
> in an autonomous car. Pay attention - this is the future!
2D thinking... what about bridge jumping ?
Would any EMP enters the Faraday cage of the car ?
It seems there is a dead zone on either sides, rear... exactly where the
current French motor-bikers go when passing you (violating the
requirement/law of 1m between vehicles)... Seems a good news, that
should provide more organs to the hospitals... Motorbike detected on the
back side... disappears... ignored... collided. Perfect.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> 2D thinking... what about bridge jumping ?
I imagine the amount of extra hardware and software required to track
objects in 3D would be immense for not much benefit.
> Would any EMP enters the Faraday cage of the car ?
>
> It seems there is a dead zone on either sides, rear... exactly where the
> current French motor-bikers go when passing you (violating the
> requirement/law of 1m between vehicles)...
Not sure, the diagram shows a dead-zone but in the text and table it
states there is a 90 degree rangefinder in the rear side doors that is
not shown in the diagram.
> Seems a good news, that
> should provide more organs to the hospitals... Motorbike detected on the
> back side... disappears... ignored... collided. Perfect.
It seems like the algorithms are sophisticated enough not to just ignore
the fact that an object has vanished from the sensors momentarily - so
you'll have to live with the crazy motorcylists for a bit longer :-)
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Le 09/01/2013 14:01, scott nous fit lire :
>
> Not sure, the diagram shows a dead-zone but in the text and table it
> states there is a 90 degree rangefinder in the rear side doors that is
> not shown in the diagram.
I'm more thinking about the location at 4:00 & 8:00 on clock, (where
corner of car are at 4:30 & 7:30)... just have the child opens the
third-fourth door... if it is only 90 degree, they are too small and a
dead zone is going to exist. Maybe not right in front of the rear side
door, but could be at the rear fuel-filling level... or better (?) at
the front side door level, right in front of the current mirror.
I can preview interesting situation with adjustment of direction & speed
to handle 1 car on the next lane at the same speed (side by side) and 2
motorbikes, both passing... one between both cars and one on the
outside. And to make things better, add one car in front and one car or
one motorbike behind, on the same lane.
When the distance to the other "car" get reduced below security level
(as both motorbikes are indeed passing you):
* move out of lane to raise it again ? (1)
* break ? (2)
* do nothing ? (3)
* other ? (4)
For additional fun, replace the back car with a heavy trunk, in hurry
(so not respecting its security breaking distance with your car).
I'm afraid automated driving is not for the Mediterranean.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> * move out of lane to raise it again ? (1)
> * break ? (2)
> * do nothing ? (3)
> * other ? (4)
>
> For additional fun, replace the back car with a heavy trunk, in hurry
> (so not respecting its security breaking distance with your car).
As I understand it the car would slow down at the same rate as the car
ahead to maintain the required safety gap. What isn't clear from the
paper is if it reacts to vehicles behind, obviously in this situation it
would be beneficial to sacrifice some of the safety gap ahead to avoid a
rear-end collision.
But you highlight one of the biggest issues for the car makers; if there
is an accident involving normal cars then one or more of the drivers
gets blamed and must pay for it, with an autonomous car the car maker
itself could get the blame.
> I'm afraid automated driving is not for the Mediterranean.
You need to weigh up how many accidents it will avoid vs how many
additional it will create. There's no reason why development should slow
down on such vehicles, and surely they will get to a point where it's
beneficial overall (I don't know if we're there yet or not).
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 09/01/2013 10:56 AM, scott wrote:
> Lots of technical detail about how the sensor and control systems work
> in an autonomous car. Pay attention - this is the future!
Lots of people are doing interesting work on AI, but the question I
always have to ask is "why bother with a multi-billion dollar computer
system when a human being can already do the exact same task for a few
pence per hour?"
In this instance, I imagine this and similar designs will never be
popular due to safety factors. The probability of an electronic system
shorting out might be roughly the same as the probability of a human
driver suddenly having an episode due to a previously undiagnosed
medical condition, but since when does probability have any baring on
human decision-making?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Am 09.01.2013 16:31, schrieb scott:
> You need to weigh up how many accidents it will avoid vs how many
> additional it will create. There's no reason why development should slow
> down on such vehicles, and surely they will get to a point where it's
> beneficial overall (I don't know if we're there yet or not).
There yet? Far from it!
They called it "urban challenge", but "small-town suburban challenge"
seems to be closer to the mark. With the trunk cram-packed with
computers, and yet still needing to be told beforehand where the roads
and lanes are supposed to be.
Yes, the car that can navigate its way autonomously from one end of the
town to the other through dense city streets without endangering or even
annoying other traffic, without driving a complete server rack around,
seems to be within a decade's reach.
It will take tremendously more effort though to build a car that will
slam the brakes hard because there's a ball rolling across the road from
between parked cars (anticipating a child to follow), yet for a raccoon
will slow down just enough to stay clear of the furball (allowing the
following traffic plenty of time to react as well).
I suspect that sophisticated computers will prove valuable to prevent
drivers from doing stupid things, but that experienced human drivers
will continue to prove invaluable to do smart things.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Am 09.01.2013 19:37, schrieb Orchid Win7 v1:
> On 09/01/2013 10:56 AM, scott wrote:
>> Lots of technical detail about how the sensor and control systems work
>> in an autonomous car. Pay attention - this is the future!
>
> Lots of people are doing interesting work on AI, but the question I
> always have to ask is "why bother with a multi-billion dollar computer
> system when a human being can already do the exact same task for a few
> pence per hour?"
Does the "DARPA" in "2007 DARPA Urban Challenge" ring any bell in you?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Wed 09/01/13 19:36, clipka wrote:
> Am 09.01.2013 16:31, schrieb scott:
>
>> You need to weigh up how many accidents it will avoid vs how many
>> additional it will create. There's no reason why development should slow
>> down on such vehicles, and surely they will get to a point where it's
>> beneficial overall (I don't know if we're there yet or not).
>
> There yet? Far from it!
I don't have any data, but my guess is the sort of car in that paper
would prevent a lot of common accidents scenarios. My point was whether
that would outweigh any additional accidents it caused that a human
usually would be able to avoid. Obviously the car will need to do a lot
better than that before it's generally accepted.
> They called it "urban challenge", but "small-town suburban challenge"
> seems to be closer to the mark. With the trunk cram-packed with
> computers, and yet still needing to be told beforehand where the roads
> and lanes are supposed to be.
More recently Google's car has done better (the one with the driving
license) but I couldn't find any detailed technical information on that one.
> I suspect that sophisticated computers will prove valuable to prevent
> drivers from doing stupid things, but that experienced human drivers
> will continue to prove invaluable to do smart things.
We are already on this path, I think computers will provide more and
more assistance to the driver but it will still be a long time before no
qualified driver is needed to oversee the computer.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> Lots of people are doing interesting work on AI, but the question I
> always have to ask is "why bother with a multi-billion dollar computer
> system when a human being can already do the exact same task for a few
> pence per hour?"
Because humans make mistakes all the time, and occasionally this causes
people to die. If there's a possibility that a computer can cause less
people to die then it's worth looking into.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
