That might be the only thing I remember from that movie. 
The wonderful thing is that you could really see a level 5 AI, trained by observation of human drivers, coming to the same conclusion! 
Not sure what standard Musk is using but it’s ridiculous to think they could achieve the SAE definition of level 5 by the end of the year: “No human intervention is required at all. An example would be a robotic vehicle that works on all kinds of surfaces, all over the world, all year around, in all weather conditions.”
That would require AI as intelligent and adaptable as a human, and if that becomes possible then the AI probably will want more out of life than driving people around.
No kidding - according to the SAE definition, the jump from level 4 to level 5 is gigantic. Fully automated on all surfaces and all weather conditions that can be managed by a human driver? I’ll believe it when I see it.
The standard that he is charging $199/month for “full self driving”.
https://electrek.co/2021/07/16/tesla-launches-full-self-driving-subscription-package-199-per-month/
So, in essence, you have to pay your car to chauffeur you around?
It even gets worse from there. Right now Teslas are pretty good at identifying relevant objects. They see cars, trucks, and motorcycles pretty well, and place them correctly (bicycle? good luck with that!). They also can do traffic cones and trash cans. Sometimes a bush is a trash can, but that doesn’t really matter as the car knows not to drive into it, and the exact nature of the object is less important.
What is really hard is making decisions once all of the objects are known. Change lanes to get around that slow car/truck object, or does the traffic pattern ahead mean staying behind the slow object is actually better? Move to the other lane now, even though there are no objects in the current lane, because the upcoming entrance ramp is going to slow things down? Change lanes to get out from behind the truck object that is going to slow down climbing the upcoming hill? Not to mention all of the defensive driving stuff (avoid that car that is behaving erratically, etc.)
In good news, this video of the new Tesla 9.1 beta is pretty impressive. It seems to be doing very well handling residential and surface streets. They talk about one of the new features being the “California stop” where it rolls stop signs. The video may be giving a bit of a deceptive view of that. At some of the stops the speedometer only gets down to 2 mph, and then the car proceeds. In my car, the speedometer may still read 2 even when the car is stopped, but not yet settled.
What I mean, is the difference between the full on “come to a complete stop and wait a half second because there’s a cop behind me”, “come very near to a full stop, and maybe the tires even stop rolling for the briefest instant, and then go”, and “‘S-T-O-P’ spells ‘yield’, roll right through”. The Tesla will still show 1-2mph for the second instance.
So, it’ll be in pretty much the same boat as the rest of us.
Telsa under investigation for hitting emergency vehicles.
Imagine you’re working at a accident scene and a Telsa approaches with a sleeping driver. The driver assist software didn’t see a fire engine stopped on the road. 
Something that was assumed in the IVHS program was that ‘car to car’ communication would precede AI driving. That would be AI communication, not accessible to the driver/passenger.
A digital comm link would solve most of the problems associated with negotiating lanes and avoiding emergency vehicles.
I think this (not being able to see stopped vehicles) is a problem inherent to relying on radar for any sort of traffic-aware cruise control (including Tesla’s autopilot) - it’s not a unique problem for Tesla.
It’s actually more of a problem with pattern recognition when relying solely on cameras. Radar can do a better job of detecting a stopped object without having to identify what it is. Lidar is by far the best at this, though.
I don’t think anything in the slate article is contrary to what I said. Radar can identify stopped objects, but it also identifies a coke can on the road, or a highway overpass as a “stopped object”, so it needs confirmation from other sensors (camera, lidar) to distinguish a parked firetruck from the background clutter, or else you end up with a lot of false-positive “phantom braking” events on the highway if you just trust the radar data alone.
AFAIK, the easiest approach is to pretty much just ignore radar returns for any object with a velocity of 0, at least whenever the vehicle’s speed is above a certain value where “phantom breaking” starts to become dangerous/scary/completely unacceptable to the consumer. At that point, the system either relies on vision (not quite there yet), lidar (expensive), or just puts all responsibility on the driver to pay attention and take over.
This isn’t a unique problem for Tesla. Volvo’s Pilot Assist, Cadillac’s Super Cruise, and BMW’s Driving Assistant all have warnings in their literature that their systems will not work when approaching stationary/stopped vehicles.
Truly self driving cars may soon move on from sitting alongside fusion power and people on Mars in the perennial “coming ten years from now” box (wouldn’t it be a trip if people on Mars won that race). But some of their developers don’t help themselves with the bluster and puffery, nor some “early adopters” with their stupidity.
I mean, to the layperson out there it must feel like, hey, there’s got to be, and if not there has to be established already, some sort of authority that can judge when it actually is happening, or at least force the company to stop calling it “Full Self Driving” or “Autopilot” when it is neither of those things, just a very fancy automated cruise control.
I have never had a car with parking assist. How well does it perform parallel parking? I’d be nervous it would hit the other parked cars or slam into the curb.
Is parking assist reliable?
It works very well on my 2015 car. It’s great fun to use when the space is barely big enough and I have a passenger.
There are some caveats though: It can have a problem with trucks that are high off the ground and it needs an identifiable kerb, not just a change of surface as sometimes happens.
You mean, when you do not have a passenger. A passenger can stand outside and tell you how far to back up, so in that case you don’t need any additional sensors.
I hadn’t considered curbless streets. I have seen them. Maybe my next car will have parking assist. Thanks bob
Here are some curbless streets, with a very high penalty for failure.
Picture shows parallel parking along a canal in Amsterdam, with no barrier between the cars and the water.