Autonomous driving

laklak · **#61** by **laklak** » Nov 29, 2016 10:24 pm

At least its not dead people with their face eaten off.

PensivePenny · **#62** by **PensivePenny** » Dec 06, 2016 12:48 pm

tuco wrote:At around 0:53, when it passes pedestrians on sidewalk, I am not sure the reaction was appropriate ideal.

Maybe you'd be more comfortable whizzing by a pedestrian walking a couple dogs at 1:23? I'm not sure which I find more disturbing. The picture isn't clear due to the frame skipping, but if you do a frame by frame it appears the autopilot doesn't even see the pedestrian.

tuco · **#63** by **tuco** » Dec 29, 2016 2:55 pm

Tesla's Autopilot predicts a crash freakishly early

In the video, which was shared online by Twitter user Hans Noordsij, you can see traffic moving along at a brisk pace on a two-lane highway. According to the dashcam readout, the cruising speed was 113 km/hr (70 mph).

All of a sudden, an alert can be heard in the car, which is Autopilot's Forward Collision Warning system warning that it has detected a potential impact up ahead.

http://www.sciencealert.com/watch-tesla ... shly-early

---

Drive safely, everybody!

I am not sure I understand this line correctly ;)

crank · **#64** by **crank** » Dec 29, 2016 7:31 pm

I think they should have that required in most vehicles, if not all of them. It's probably expensive, but it needn't be, the components aren't. Spread the development over millions of vehicles and it would likely cost less than a electric seats, maybe less than a set of airbags..

tuco · **#65** by **tuco** » Dec 29, 2016 7:44 pm

Since the vid included in the article was taken down, here its for your convenience:

I am still confident that insurance companies will put enough pressure on unbelievers sooner or later. With help of legislature of course.

OlivierK · **#66** by **OlivierK** » Dec 30, 2016 12:16 am

That's a good catch by the autopilot, but there are two things I wonder:

What's it's rate of false positives? I'd imagine it would err on the side alerting to possibly dangerous situations even if they don't go on to become accidents. In which case it would be a worry if there was an accident that it didn't see coming.

Also, is this necessarily any better than human risk assessment? Most humans can also see crash situations unfolding before they hear the bang.

So yeah, it's impressive to see this level of automated response. Is it evidence of superiority of automated systems? Maybe. Maybe not.

Macdoc · **#67** by **Macdoc** » Dec 30, 2016 12:39 am

The insurance companies think the gains will far outweigh the few negatives but only some real world tests will prove the concept and I think agressive drivers and motorcycle riders will be penalized by slower moving traffic in lockstep.

crank · **#68** by **crank** » Dec 30, 2016 3:38 am

OlivierK wrote:That's a good catch by the autopilot, but there are two things I wonder:

What's it's rate of false positives? I'd imagine it would err on the side alerting to possibly dangerous situations even if they don't go on to become accidents. In which case it would be a worry if there was an accident that it didn't see coming.

Also, is this necessarily any better than human risk assessment? Most humans can also see crash situations unfolding before they hear the bang.

So yeah, it's impressive to see this level of automated response. Is it evidence of superiority of automated systems? Maybe. Maybe not.

We have false positives all the time, well, I do, I assume it happens with other people, seeing what looks like it might develop into an accident and slow down etc. I guess I should say it better happen with everyone, because no one is so good that they won't slow down unless the accident actually happens.

I would think, as maybe happened here, that getting the 3D data you could glean from the radar would allow the system to see things we couldn't because the vehicles are almost in line--in the same direction and we can't judge their separation along the line of sight very well at all. A car that would appear to be changing lanes safely could actually be about to hit a car in the next lane, we likely thinking there was room, else why would they change lanes?

tuco · **#69** by **tuco** » Dec 30, 2016 6:53 am

OlivierK wrote:That's a good catch by the autopilot, but there are two things I wonder:

What's it's rate of false positives? I'd imagine it would err on the side alerting to possibly dangerous situations even if they don't go on to become accidents. In which case it would be a worry if there was an accident that it didn't see coming.

Also, is this necessarily any better than human risk assessment? Most humans can also see crash situations unfolding before they hear the bang.

So yeah, it's impressive to see this level of automated response. Is it evidence of superiority of automated systems? Maybe. Maybe not.

If nothing else, AI pays attention all the time so it is necessarily superior.

Besides, have you read the article? Because, and we've been over this already, human sensors and computing power are inferior. I do not want to hear about roos again, ok?

On a lighter note:

Sendraks · **#70** by **Sendraks** » Dec 30, 2016 10:10 am

tuco wrote:If nothing else, AI pays attention all the time so it is necessarily superior.

Combine this with the fact that an AI can pay attention all the time to the road, whilst simultaneously paying equal attention to navigating to a destination and not getting distracted by stuff the way the human brain can.

OlivierK · **#71** by **OlivierK** » Dec 30, 2016 10:24 am

Well, yeah, the always-on aspect (as well as being able to always be looking forward, left, right, and behind simultaneously) is superior, but it's a package deal with the interpretation of the input. Ten years ago, we could have had the sensors going 100%, and the car still wouldn't drive as well as a human, because the software wasn't good enough. Now it's better, and seems pretty line-ball with human analytical ability. Soon, it will no doubt be better still.

I'm not against self-driving cars in the slightest, but saying they've reached a superior level takes more than just noting that their sensors are always on. In particular, the video posted here doesn't strike me as particularly impressive in that regard - the system detected a moderately difficult to detect situation, and dealt with it from a pretty easy-to-deal-with-it distance.

And tuco, don't lecture me about roos, that was someone else that brought them up, and I merely pointed out that a self driving car would likely have taken exactly the same action as our heroic doubter took to save himself.

tuco · **#72** by **tuco** » Dec 30, 2016 11:04 am

The roos were joke!

It was very good example, as noted in my initial post here, of challenges AI programmers are facing. I even recognized your expertise of roos and if I was to program AI I would hire you as consultant for roo algorithm.

In the vid there are, in my opinion, two important elements: a) always-on principle b) keeping safe distance at given speed. While human drivers are probably capable of both, we would need to go through data and look up what causes most accidents to confirm my guess that a) and b) are among the most frequent in given situation.

To the point .. indeed, only sensors do not make autonomous cars superior. That is given, 11 year old stuff, no point to waste time and energy pointing it out or arguing it.

Also, is this necessarily any better than human risk assessment?

Is what? Obviously, for the time being, its not better yet since its work in progress. Assuming AI will get better, which is a reasonable assumption because driving is not all that complex task and what makes it dangerous is mostly speed/reaction time, IT is superior because of sensors.

Have you seen any roaches bumping to each other or falling of a cliff or something? IT is not as complex as IT might seem.

Sendraks · **#73** by **Sendraks** » Dec 30, 2016 11:15 am

tuco wrote:
In the vid there are, in my opinion, two important elements: a) always-on principle b) keeping save distance at given speed. While human drivers are probably capable of both, we would need to go through data and look up what causes most accidents to confirm my guess that a) and b) are among the most frequent in given situation.

I had a quick look online and a summary of the top causes of road accidents in UK are:

•Speeding.
•Drink-driving.
•Not wearing a seat belt.
•Careless or aggressive driving.
•Drivers who don't look.

All of which are incidents that AI control would mitigate against and in a scenario where all cars were AI control, would eliminate entirely.

In addition, AI vehicles obeying the speed limit, not rushing through traffic lights or doing other stupid human stuff like jumping junctions, would improve the flow of traffic overall resulting in a smoother commuting experience and shorter journey times. Pretty much all my lengthy road commutes have been a result of an error being committed by a human driver somewhere in the system.

tuco · **#74** by **tuco** » Dec 30, 2016 11:33 am

I think hope we agree that general driving is not the problem. The problem are unpredictable elements, where yet again sensors will help to reduce them. The thing with unpredictable situation is that drivers have to make split second decision so its doubtful if there is any "risk assessment" at work at all. So, ironically, johnbrandt earlier post makes a good point. Its however not problem of AI not being able to "compute" it.

Sendraks · **#75** by **Sendraks** » Dec 30, 2016 12:00 pm

tuco wrote:I think hope we agree that general driving is not the problem. The problem are unpredictable elements, where yet again sensors will help to reduce them. The thing with unpredictable situation is that drivers have to make split second decision so its doubtful if there is any "risk assessment" at work at all. So, ironically, johnbrandt earlier post makes a good point. Its however not problem of AI not being able to "compute" it.

I think that, generally, the overall increase in safer driving that AI controlled cars would result in would reduce the occurrence of these unpredictable situations and where they did occur, safer driving would mean less disastrous options than john suggests.

Macdoc · **#76** by **Macdoc** » Dec 30, 2016 7:30 pm

•Not wearing a seat belt.

might be a cause of fatality - not a cause of the accident

laklak · **#77** by **laklak** » Dec 30, 2016 8:36 pm

There is almost no such thing as an "accident", it's somebody's fault. Somebody didn't look when changing lanes, somebody was testing, somebody was having a wank. We need to find this Somebody person and beat them senseless. About the only real accident is equipment failure, in pretty much every other case it's human error. From that standpoint AI would be a distinct improvement.

Hell, given how most of the fossils around here drive, intelligence of any sort, artificial or otherwise, would help.

OlivierK · **#78** by **OlivierK** » Dec 30, 2016 9:10 pm

That's a good point. Having just done over 1000km visiting family for Christmas, my main traffic issues were slow people passing other slow people on motorways. Speeding up the fossils would be an improvement. Of course, the biggest issue around holiday time is the grey nomads towing caravans at 80kmh on motorways because that gives better fuel economy, and they've got all the time in the world (but not enough to take the slower highway that runs parallel, because that's got junctions that would screw their mileage.)

Also, autonomous cars might be a bit faster than the tourists who come down our road at about half the speed we drive it because they've got no experience with the road (fair enough) or indeed any narrow rural road (annoying). If we get good algorithms, then I hope that one aspect is if a vehicle is travelling slower than others for some valid reason (towing, small engine), then the algorithm knows to pull over and let others pass where overtaking lanes aren't available.

tuco · **#79** by **tuco** » Dec 30, 2016 9:27 pm

I was listening to some expert on this, someone involved in R&D, and he said something like .. some people oppose it because they do not want to give up their driving. I understand that, but sorry bro I had to give up smoking around other people.

Macdoc · **#80** by **Macdoc** » Jan 03, 2017 2:25 am

MIT study says 3,000 ride-sharing cars could replace every cab in New York City
by Jordan Golson@jlgolson
All 13,000 taxis in New York City could be replaced by a fleet of 3,000 ride-sharing cars if used exclusively for carpooling, according to research published today by MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). Instead...

http://www.theverge.com/2017/1/2/141472 ... -uber-lyft

Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Re: Autonomous driving

Who is online