The lane markers in this area, based on video I watched of it years ago, are pretty bad, IIRC. Like, a human could have problems following the markers bad. As evidence to support this I'll say that there had been an accident in this exact area days (again, IIRC) before that destroyed the "crumple zone" in this area. :-)
It's going to be extremely interesting to see how the courts rule in a case where "I knew the Tesla on autopilot would drive in to the barrier in this area, and I turned on autopilot in this area and it drove into the barrier."
- Autopilot doesn’t work in that area. Driver reports it to Tesla.
- Tesla fixes the issue in a software update.
- Driver sees the improvement, and happily relies on autopilot.
- After a while, a new software update introduces a regression.
- Driver doesn’t find out about the regression until it’s too late (and is not around to testify about it anymore).
Regressions undoubtedly happen, but clearly in the vast, vast majority of cases, no deaths are involved. It's not "the problem," it's "the exception."
[1] https://en.wikipedia.org/wiki/List_of_self-driving_car_fatal...
It would be interesting to design the ux for this. Based on the wording it could even make them more liable if accidents happen outside of roads with known issues.
Sounds like a sensible idea, particularly in an area where they've seen a near collision. The difficulty might be that conditions change all the time (barriers are moved or removed), and that makes it difficult to make this kind of determination on a case by case basis.
You just learned to live with the limitations and used it where it worked under supervision.
Seems unlikely, but I just don't understand why he'd be using Autopilot in this area given his past with it.
When I first got Autopilot in my Tesla, I was playing with it on the city streets, and it decided to swerve into the divided, concrete median at the end of the intersection, deciding that was the lane. I grabbed control of it, no problem, but 2 years later I still am extra vigilant around any lane dividers. Mostly though, I don't use Autopilot in such restricted areas.
The black-and-white stripes on the highway divider were good, though[1]. They are hard for humans to avoid.
That's the fundamental problem: the signage optimized for being glaring to humans is sometimes invisible to self-driving cars.
Instead of trying to teach cars to read signs not meant for them, we should install signs that are easy for cars to understand.
[1]https://www.mercurynews.com/2018/03/28/tesla-claims-highway-...
(this also shows the lack of crumple zone, but clear signage).
> The grim statistics provided by the National Highway Traffic Safety Administration also show that drivers on rural roads die at a rate 2.5 times higher per mile traveled than on urban highways. Urban drivers travel twice as many miles but suffer close to half the fatal accidents. [1]
Could this explain away Musk's AP safety claims, given that it's mostly used on highways?
[1] https://www.npr.org/2009/11/29/120716625/the-deadliest-roads...
https://www.thedrive.com/tech/26455/nhtsas-flawed-autopilot-...
"Anyone who relies on LiDAR is doomed. Doomed. It's like having a whole bunch of expensive appendices. One is bad, and now you want to put in a whole bunch of them. That's ridiculous."
Companies using LIDAR? "They bet wrong".
People describing how stereo cameras don't inherently see 3D? "Baloney. Based on eyesight".
Companies that have spent years and billions developing and testing autonomy? "Almost certain they are wrong".
It's a little more compelling on the data side but also goes into Lidar vs vision.
It might be controversial at this point to assert that vision is better. But I don't think it's controversial any longer to assert that vision is sufficient.
Given that driving and roads are currently set up for vision and Lidar will never be able to "see", it seems very reasonable to think vision could ultimately be better.
"Sure it’s laggier but with dedicated silicon that’s a non-issue.
Fact is vision is more than adequate. And lidar is literally blind."
There's no evidence that I have seen that suggests vision-based autonomy will not be sufficient. Tesla is not the only one using the approach. Lidar has its own problem with cost and that it literally cannot see.
On the other hand, humans do not have lidar systems in their head, and so lidar is the real unproven technology. It's also clear that lidar is especially lacking in situations which require seeing longer distances (like say, driving!)
To be fair, lidar is winning in terms of problems vs. miles driven, but I believe their advancement will plateau far sooner (and perhaps permanently) than computer vision based systems.
We'll find out in a few years...
Also, if the pilots input a heading change, the plane will execute a 1G turn using all control surfaces so it's almost imperceptible to the passengers.
I think you're oversimplifying airliner autopilot.. by a lot.
There is certainly no such thing as an auto-land system in which the pilots are not required to be ready to take full control of the aircraft at a moment's notice. It is Level 2 in the strictest sense.
The difficult part is decision-making in non-standard situations like Sully landing on the Hudson with his A320.
There is a growing body of videos from dashcams (a feature enabled by another software update) on youtube showing Teslas avoiding collisions that would have been caused by other drivers being negligent.
But the simple autopilots (think an old Cessna) don't even try to avoid the ground for you. If you set it and forget it, you may find yourself dead on a mountainside.
And low pressure zones.
If you are interested to read more
Clearly, with their naming of their system as autopilot rather than cruise control, Tesla is trying to imply that autopilot means something more. Namely, it means the driver / pilot can offload the work of routine driving / flying to the computer, only intervening in certain times (e.g. off0highway or takeoff / landing) or in the event of failures / extreme conditions.
Cruise control only enables the latter.