Autonomous Cars

Breaking news, useful data or technical highlights or vehicles that are not meant to race. You can post commercial vehicle news or developments here.
Please post topics on racing variants in "other racing categories".
User avatar
loner
46
Joined: Fri Feb 26, 2016 5:34 pm

Re: Autonomous Cars

Post

para bellum.

User avatar
strad
269
Joined: Sat Jan 02, 2010 12:57 am

Re: Autonomous Cars

Post

AAA did a study and found autonomous braking doesn't work nearly as well as the manufacturers claim.
Very disappointing to buyers and they were advised not to count on self braking apps. #-o
To achieve anything, you must be prepared to dabble on the boundary of disaster.”
Sir Stirling Moss

User avatar
hollus
Moderator
Joined: Sun Mar 29, 2009 12:21 am
Location: Copenhagen, Denmark

Re: Autonomous Cars

Post

"For every complex problem there is an answer that is clear, simple and wrong." H. L. Mencken
(¡Puxa Esportin! temporarily on hold due to you know what)

Greg Locock
Greg Locock
179
Joined: Fri Jun 29, 2012 11:48 pm

Re: Autonomous Cars

Post

Here's the report. Total fail by GM and Tesla at 20 mph. They saw the pedestrian but splattered him. Toyota was the best system.

https://www.aaa.com/AAA/common/aar/file ... ection.pdf

User avatar
strad
269
Joined: Sat Jan 02, 2010 12:57 am

Re: Autonomous Cars

Post

yes Hollus that's the report.
Seems that over all they failed to live up advertising and what they lead the consumer to believe.
To achieve anything, you must be prepared to dabble on the boundary of disaster.”
Sir Stirling Moss

User avatar
loner
46
Joined: Fri Feb 26, 2016 5:34 pm

Re: Autonomous Cars

Post

China started developing 6G...
para bellum.

User avatar
strad
269
Joined: Sat Jan 02, 2010 12:57 am

Re: Autonomous Cars

Post

Watched an interesting documentary the other night where in the leading experts developing AVs stated unequivocally that they themselves think total AVs (stage 4 and 5) are still a long way off. And this was from people with a lot of time and money invested in pursuing it. I was shocked myself. It wasn't from people you'd expect to say so.
To achieve anything, you must be prepared to dabble on the boundary of disaster.”
Sir Stirling Moss

User avatar
RZS10
145
Joined: Sat Dec 07, 2013 12:23 am

Re: Autonomous Cars

Post

https://www.nbcnews.com/tech/tech-news/ ... e-n1079281

I think that case had been discussed here when it happened - they did not program in that pedestrians might ... jaywalk ...

Greg Locock
Greg Locock
179
Joined: Fri Jun 29, 2012 11:48 pm

Re: Autonomous Cars

Post

Uber's software seems full of holes. One obvious one is that it doesn't build a history for an object, so it didn't see the convergence of the course of the victim and the car, but only started to react when she was in front of the car. This is such a basic error, ignoring 60 years of target tracking, and 400 years of naval navigation, it really makes you wonder what the Hell their processes are.

User avatar
Phil
412
Joined: Tue Sep 25, 2012 3:22 pm

Re: Autonomous Cars

Post

The problem is it’s zero intelligence.
Not for nothing, Rosberg's Championship is the only thing that lends credibility to Hamilton's recent success. Otherwise, he'd just be the guy who's had the best car. — bhall II
#Team44 supporter

User avatar
rscsr
68
Joined: Sun Feb 19, 2012 12:02 pm
Location: Austria

Re: Autonomous Cars

Post

Greg Locock wrote:
Sun Nov 10, 2019 2:23 am
Uber's software seems full of holes. One obvious one is that it doesn't build a history for an object, so it didn't see the convergence of the course of the victim and the car, but only started to react when she was in front of the car. This is such a basic error, ignoring 60 years of target tracking, and 400 years of naval navigation, it really makes you wonder what the Hell their processes are.
I read that they actually build a history for an object. But in that case it was reclassifying her over and over between a standing pedestrian and a cyclist. And they seem to throw away the history whenever they reclassify an object for whatever reason.

Greg Locock
Greg Locock
179
Joined: Fri Jun 29, 2012 11:48 pm

Re: Autonomous Cars

Post

That makes more sense (but not much).

User avatar
Phil
412
Joined: Tue Sep 25, 2012 3:22 pm

Re: Autonomous Cars

Post

A human doesn't need to know and have seen every object in existence to know what it might be seeing or to make the right decision while driving. Based on logical reasoning, we can make pretty good assumptions. If we see a human walking, drunk, crouched, running, crawling, we are still able to identify it as a human and not misidentify it as an animal or anything else, least because we are also able to interpret the context of where and when we are seeing it.

To do something like this in software (with zero-intelligence) is rather complex, if not to say impossible.

One thing I find rather more intriguing though - even if that AV couldn't make the proper distinction between bicycle and human - it shouldn't matter, because it was an object in a direct path plain and simple. In the end, it could have been an elephant, a robot from the future, a guy concealed in a box crawling over the road... it still should have stopped.

It's one thing to identify what you are seeing - but it's another of simply seeing that there's an object within the path of the car and no matter what it is, to stop.

If the whole AV industry is built with the goal of simply identifying anything 'moving' that could be potentially on the road, I wonder what will happen if there's an object there that a piece of software simply wasn't programmed to identify.
Not for nothing, Rosberg's Championship is the only thing that lends credibility to Hamilton's recent success. Otherwise, he'd just be the guy who's had the best car. — bhall II
#Team44 supporter

User avatar
Big Tea
82
Joined: Sun Dec 24, 2017 7:57 pm

Re: Autonomous Cars

Post

Phil wrote:
Fri Nov 15, 2019 3:08 pm
A human doesn't need to know and have seen every object in existence to know what it might be seeing or to make the right decision while driving. Based on logical reasoning, we can make pretty good assumptions. If we see a human walking, drunk, crouched, running, crawling, we are still able to identify it as a human and not misidentify it as an animal or anything else, least because we are also able to interpret the context of where and when we are seeing it.

To do something like this in software (with zero-intelligence) is rather complex, if not to say impossible.

One thing I find rather more intriguing though - even if that AV couldn't make the proper distinction between bicycle and human - it shouldn't matter, because it was an object in a direct path plain and simple. In the end, it could have been an elephant, a robot from the future, a guy concealed in a box crawling over the road... it still should have stopped.

It's one thing to identify what you are seeing - but it's another of simply seeing that there's an object within the path of the car and no matter what it is, to stop.

If the whole AV industry is built with the goal of simply identifying anything 'moving' that could be potentially on the road, I wonder what will happen if there's an object there that a piece of software simply wasn't programmed to identify.
Not on the same scale maybe but have you used google lens?
One test is worth a thousand expert opinions

User avatar
strad
269
Joined: Sat Jan 02, 2010 12:57 am

Re: Autonomous Cars

Post

In that documentary I mentioned they showed a pedestrian in a hoodie get hit by an AV because it, the AV had not been trained to recognize someone in a hoodie as a human.
To achieve anything, you must be prepared to dabble on the boundary of disaster.”
Sir Stirling Moss