Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

New Tesla AP 2.0 Self Driving video

This site may earn commission on affiliate links.
While I understand that this is just a demo of current state of the FSD, it does show a lot of mistakes the car made
As I noted in another thread discussing this video, it only shows data from 3 of the 8 cameras, doesn't show the radar data and doesn't show the data from the 12 ultrasonic sensors. So it would be premature to look at the video and decide what "mistakes" the car made since its actions are based on 21 data streams and the video only shows 3 of them.

I am not saying that the video shows that the car did that drive "perfectly". It appears that it did not. I am saying that one should be careful about deciding what was a "mistake" when one does not have all the data needed to make that judgement.
 
  • Helpful
Reactions: SW2Fiddler
So it would be premature to look at the video and decide what "mistakes" the car made since its actions are based on 21 data streams and the video only shows 3 of them.
Also, I take heart in the fact that (maybe editing?) at no point does it just stall confused. Even calling some of the actions "mistakes" aligns with human driving quality -- you make mistakes, you might even verbally emit "oops", and then you move from there to a safer location in the state machine and continue.
 
  • Like
Reactions: Krugerrand
So it would be premature to look at the video and decide what "mistakes" the car made since its actions are based on 21 data streams and the video only shows 3 of them.

I wasn't making determination of "mistakes" based on the data from the sensors, but based on a normal driving patterns of a human. That's not entirely fare, but that's the only thing you can go on based on the video. Some of those issues are minor, but stopping in the middle of the busy intersection is never a good idea, no matter what those 21 data streams are telling you.
 
I tend to agree with ivolodin's sentiment. It's the 'false positives' which are the main concern to me and the prospect of non-autonomous cars following behind and not expecting the car in front to slow suddenly for no apparent reason. The majority of drivers don't leave enough room and don't pay sufficient attention to stop in good time for a car which unexpectedly slams it's brakes on in front of them.

If all cars are equipped with the same self driving technology, then it's no big deal if they all slow or stop every now and then for no good reason, but we're soon going to have to cope with a mix of humans and computers navigating their way around an increasingly complex environment and that will lead to 'mistakes', even if you call them something different.
 
It's exciting news for my child's friend that's legally blind. She will have a chance at independence as she gets older. Super cool!

The way technology is advancing in leaps and bounds, she'll hopefully get her eyesight back.

Once people get used to riding in fully autonomous cars (and I don't think it will take very long to get used to them) they will look back in wonder at the days when failure-prone humans were allowed to control vehicles weighing thousands of pounds and moving at very high speeds with tanks full of highly flammable liquids. It will seem insane to future generations.

Yes, but it would sure be sad if I had to get up every morning and have my car drive me to work and back, and then on the weekend it drives me to my cabin and back, and all the while I have to just sit and watch it drive. If I wanted to drive for fun, I would have to go to a driving course, like people do with horses, when not too long ago horses were allowed on the road but have since been banned (unless you live in the country or a small town). Human driven cars are headed for bans too. Probably not in my lifetime, though, fortunately.
 
While I understand that this is just a demo of current state of the FSD, it does show a lot of mistakes the car made and some of them were pretty serious. At 0:31 it got into a hazard situation when taking a left turn while the car in the opposite lane was taking a right turn. While it was following the rules, it could potentially result in an accident. It did stop in the middle of the intersection, because of that. At 00:40 seconds the car randomly stopped (possibly to let the bicycle through, although it's hard to tell). At 1:31 it randomly stopped in the middle of the road, I guess because it saw pedestrians that were too close the the road, although they are clearly on the side-walk. At 1:41 it seems to overshoot it's turn and had to come to a complete stop in the middle of the intersection. At 2:33 it actually got onto the wrong side of the road while take a right turn and had to come to a complete stop in the middle of the busy intersection. And it's really hard to tell what is going on in the parking lot.
Don't get me wrong, it's still impressive what Tesla managed to accomplish in a short period of time. But it also shows that FSD is pretty far from being done.

P.S. Times are based on the longer version: Full Self-Driving Demonstration

I think you nailed it. The system can have 500 data streams, all are irrelevant if it can't drive properly. Trying to look at it from the car's perspective makes it easy to offer a typical Tesla apologist excuse and dismissal of the concern, in my humble opinion. The car's perspective is totally irrelevant, what matters is what is occurring from the driver's perspective. The part that jumped out at me was the car stopping unexpectedly because of the pedestrians on the side of the road. That, alone, could be enough to get rear-ended.

The video is impressive at first, but when you look at the details, there is clearly much work yet to be done. Not recognizing animals is a major, major issue that I hope is addressed. What will the car do if a dog runs out into the road? Will it brake severely to save the dog, or if there is a car following closely behind, will it just hit the dog to avoid being rear-ended? I love technology and all, but these technologies raise more questions than they solve problems, in my opinion.

I'm a bit more cynical than others here. I don't think Tesla is developing this technology at such a breakneck pace because of safety and out of a sense of altruism. I believe their driving motivation is to create a differentiating feature ahead of what is likely to be a large influx of competitors in the next few years. Once Audi and others have EVs with equivalent range to Teslas and with likely much higher quality and fit/finish, there will really be very little reason to consider a Tesla unless it is much cheaper or offers something the others don't.
 
Last edited:
What will the car do if a dog runs out into the road? Will it brake severely to save the dog, or if there is a car following closely behind, will it just hit the dog to avoid being rear-ended? I love technology and all, but these technologies raise more questions than they solve problems, in my opinion.

It will make a better decision than 99.99% of humans, in my view.
 
To me the driving looked pretty typical of Palo Alto and the Bay Area in general. Stopping for no apparent reason? Check. Driving well under the speed limit? Check. Hesitating at a turn when it had right of way? Check. I think the AI is just trying to fit in.

Anyway this is just another demo. I'm confident that Tesla's public release will make reasonable efforts to avoid running over dogs etc. But the sad truth is that human drivers do things like that every day. I'll be satisfied if we can show that the HAV software is substantially safer, on average and for most outlier cases, than human drivers are.
 
Yes, there were a few driving glitches. None caused problems on the road (the car never stopped in the middle of an intersection). Animal detection does need to be improved.

Nonetheless, this is a very, very impressive video. You can clearly see how they will be able to achieve good urban autonomous driving from this point. It is, as Elon promised, being implemented much faster than we had all thought possible. No wonder they ditched Mobileye.
 
Once people get used to riding in fully autonomous cars (and I don't think it will take very long to get used to them) they will look back in wonder at the days when failure-prone humans were allowed to control vehicles weighing thousands of pounds and moving at very high speeds with tanks full of highly flammable liquids. It will seem insane to future generations.
read some Larry Niven stories of the not so distant future
 
  • Helpful
Reactions: SW2Fiddler
The part that jumped out at me was the car stopping unexpectedly because of the pedestrians on the side of the road. That, alone, could be enough to get rear-ended.
Unless the cars are all crash avoidance enabled to avoid rear end collisions
My friend caused a wreck, because the driver was watching her walk
my uncle Charles was run over and killed at age 7 while on a bicycle, because the chauffeur was "distracted" by a pretty girl in 1917
I suspect you worry too much. The future is approaching faster than you realize or are ready for
 
Not really. I shake my head almost every day watching humans drive. I doubt I'd be doing that if AP2.0 was driving. It won't be elderly, partially blind, suffer from road rage, had a few or more drinks, on medication, and the list goes on and on.

I've had some uber drivers that I seriously did not feel safe riding in their car. Last minute turns, crossing lots of lanes of traffic. Last min stopping, tailgating, etc.
 
  • Like
Reactions: wesley888
interesting time points:

0:43 cyclist drives in front of the car without stopping at the stop sign

90%+ cyclists never stop at stop signs.

I would like to see Tesla release a normal-speed video, I don't care if it's 15-20 minutes long. This sped-up stuff is a cheat.

I agree with @ivolodin that the video shows the car made mistakes, which, I'd argue, would be more obvious if the video ran at normal speed.

Finally, I continue to be concerned that Tesla's release of these kinds of videos, while fascinating to techies and early adopters, could be misleading to mainstream average viewers not as fully versed in all things Tesla and the state of the autonomous vehicle technology art. I think it is rather irresponsible of Tesla to release videos that have no explanations, annotations, introductions, or other proper context-setting so the public in general not to mention the less clueful media absorb and interpret the video correctly. I can imagine people viewing this video thinking this is what a brand-new Tesla can do right now. I can even imagine a non-sophisticated new owner thinking this.