Eng-Tips is the largest engineering community on the Internet

Intelligent Work Forums for Engineering Professionals

Tesla "autopilot" disengages shortly before impact 9

Status
Not open for further replies.

MartinLe

Civil/Environmental
Oct 12, 2012
394
0
0
DE

"On Thursday, NHTSA said it had discovered in 16 separate instances when this occurred that Autopilot “aborted vehicle control less than one second prior to the first impact,” suggesting the driver was not prepared to assume full control over the vehicle."

Where I a legislating body, I would very much demand that safety critical software needs to be as transparent as any other safety feature. Which would limit "AI"/machine learning applications in these roles.
 
Replies continue below

Recommended for you

Spalso, I too know a driver that lost his vision (in a crash of all places). He has safely navigated familiar roads for the last decade and can ride small motorcycles but not comfortable getting back on freeway capable sized bikes.

Something very important to consider is your brain's ability to learn for decades and then function using that retained knowledge despite losing an input. I think learning to drive without depth perception would be much more challenging than losing your depth perception after learning how to drive.
 
I don't think it's just Tesla.

As I've said a few times on these forums, I love my adaptive cruise control which uses a small lowdown radar in the front grill.

HOWEVER, I did start to notice that it must be quite a narrow beam because when following a motorcycle which moved out from the centre of the lane, my car decided it wasn't there any more and started accelerating (!)

So now following motor cycles I turn it off as it does seem to only find what is in the middle third of the lane.

So be aware.

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.
 
For non-motorcyclists, FYI, motorcyclists generally travel in the left tire track of the lane. In a group of two or more, they typically alternate between the left and right tire tracks, while spacing themselves longitudinally as well.

"Schiefgehen wird, was schiefgehen kann" - das Murphygesetz
 
I guess the idea of spacing several radars out front and integrating the outputs is way out of line.

It's all about cost, possibly at the expense of safety. Tesla's approach, I think, is based on that; Musk eschews lidars because of cost, but rationalizes it by claiming that human-like vision is all that's needed. Far be it for him to admit that he compromised on safety because of cost.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
Maybe we should let Elon go his own way on this, but not let him sell it to others until it actually works. Those other manufacturers who care to add lidar, radar, zip-ties and cowcatchers (or whatever) so as to make THEIRS work can go ahead and sell theirs.

Thus they will be making sales, and money, long before Elon does. And Elon will be proven right. Eventually. I'm sure he will be very happy to prove he was right after all.

And hardly anyone dies as a result.

I like it.


spsalso
 
It's possibly patent related. Nvidia put their foot in the door early using their GPUs to do image processing in reverse or what they do in video games and such. Perhaps lidar requires extensive image processing of which Nvidia holds parents?
 
TugboatEng,

I spent thirty five years working for a LiDAR company. Software development is a huge part of what they do. All the LiDAR has to do is see an object and work out its velocity and acceleration. There are no objects the robot vehicle can safely smash into.

--
JHG
 
All the LiDAR has to do is see an object and work out its velocity and acceleration.

I've been beating that dead horse for a long time, here, but the algorithm guys seem hellbent on trying to identify the objects, presumably for behavioral predictions, but they (Tesla) seemingly have no fallback position if the detected objects aren't the same, and they don't seem to have entertained the notion of time history and memory.

Perhaps lidar requires extensive image processing of which Nvidia holds parents?

Nvidia is a merchant company that sells both hardware and IP, which nullifies any patent concerns. It's more likely a liability issue; Nvidia does not sell these as mission-critical items, since that involves reliability analysis above and beyond anything they might want to address.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
I don't know how the system works ... BUT regardless ... Today, the forward-collision-warning system on my Chevrolet Bolt picked up on the fact that a driver up ahead had slammed on the brakes and was slowing rapidly even though they were still a decent distance ahead. The visible and audible warning popped up while my foot was in the process of going from the accelerator to the brake pedal, so its perception of what was happening was in agreement with my own.

This was the first time this system had ever triggered, and it was for a real event that warranted taking action, not a false-trigger.
 
IRstuff said:
I've been beating that dead horse for a long time, here, but the algorithm guys seem hellbent on trying to identify the objects...

robot said:
Hey, that's little Billy of 123 Elm Street. If I go up on the sidewalk to hit him, there is a good chance I will kill my two passengers. On the other hand, he has two overdue library books, and we need to make an example of the little bastard.

If the robot sees an object that is not on its internal map, the most likely explanation is that it is a mobile object. Even if the object is not moving and is evidently mot capable of movement, there is no guarantee that something will not step out from behind it. The robot must take evasive action by slowing down and pulling across to maintain distance. This is all so simple.

--
JHG
 
Yup, so simple.

Since a kid can run out between two parked cars, the robot must stay far enough away from parked cars so as not to hit any child who might run out.

That pretty much eliminates any lane adjacent to parked cars. Unless the vehicle is traveling at maybe 3 MPH.


spsalso
 
spsalso,

I live in a city that is old by North American standards, and there are lots of narrow streets with parked cars. I try to stay off those streets, and if I can't, I drive slowly (25mph/40kph) and I keep my distance from the parked cars. Robots need to do this.

--
JHG
 
You drive defensively, one of the expectations with AI is that it does not bc most people do not lest it be seen as inefficient. I've wondered for several years if that isn't the same reason for their illegal behavior/programming.
 
The robot must take evasive action by slowing down and pulling across to maintain distance. This is all so simple.

I'm not disagreeing, but someone needed to tell Uber's software gurus that little tidbit, and they possibly could have avoided killing a pedestrian on a nearly empty roadway.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
CWB1,

If I design a robot car, I am the guy who gets sued if it takes out a troop of girl guides. The robot will drive defensively. You won't like driving behind it.

--
JHG
 
Since a kid can run out between two parked cars, the robot must stay far enough away from parked cars so as not to hit any child who might run out.

There are going to be unavoidable things, this might be one; nevertheless, all the sensors on a self-driving car are operational 24/7, so those accidents that occur because the human driver had lapsed attention might possibly be avoided.

And lidar potentially can detect much more than a human driver, partly because of where the sensor is mounted, rooftop provides a slightly better vantage, while a bumper mount could potentially see under cars

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
"I'm not disagreeing, but someone needed to tell Uber's software gurus that little tidbit, and they possibly could have avoided killing a pedestrian on a nearly empty roadway."

Do not hit something larger than a breadbox (old cultural reference).

"If I design a robot car, I am the guy who gets sued if it takes out a troop of girl guides. The robot will drive defensively. You won't like driving behind it."

Do not hit something larger than a breadbox.

DO NOT HIT SOMETHING LARGER THAN A BREADBOX.

IF something larger than a breadbox suddenly appears in your (robot's) perception, do the best you (robot's) can.


I will note that earlier in this discussion, an item larger than a breadbox appeared in front of a vehicle. This was a stopped tractor trailer. The vehicle hit it.

BIG FAIL!


spsalso

 
Status
Not open for further replies.
Back
Top