Self driving cars are still decades away

Yes and no. It’s a fairly straightforward edge case, so it should be relatively easy to train.

But none of this is easy. There isn’t some big IF/THEN/ELSE statement where you just insert the school bus rule. When you train it for a stopped school bus, what else have you introduced into its logic? What else does it think “looks” like a school bus? What other cases that used to work now maybe don’t?

To me, it seems like such an obvious case, which suggests a couple possibilities. Either it does work and this report is bogus, or it’s not as easy as it seems.

The standard school bus when it stops for kids has a stop sign that pops out from the side and a bunch of lights that go crazy and they are a certain color yellow and have the same signage. At least around here.

In Ohio, the stop sign isnt mandatory. It’s been a long time since I’ve seen a bus without the sign, but technically any bus stopped with lights flashing needs to be treated the same.

Not sure how other states differ.

You just need to make sure they can’t prove it was you who torched it–[for example the cameras just show the bunch of protesters stopping the vehicle’s progress]. It’s the same type thing where you order an expensive item from Amazon and say you didn’t get it–it must have been stolen off your porch–while you are the one who stole it. That is called friendly fraud and makes up the majority of credit card fraud:

While off-topic, and I don’t recall which video it was, I heard a comment that a possible explanation for attacking the Waymo taxis is that some people believe that the video captured by their on-board cameras can be/has been used by police or other government agencies for surveillance during protests. I don’t know if this happens, but it is a possible motive for the torchings.

Off topic, but the more pedestrian explanation is that human drivers avoided the riot but the Waymos happily followed their route. Targets of opportunity + mob logic.

This is key. Every state has slightly different standards for what a stopped school bus looks like.

Human-level cognition can deal with the vagaries of a stopped bus in an unfamiliar state that e.g. paints them pink. Computer level cognition? Not yet. At least not yet reliably.

Tesla Robotaxis will supposedly roll out soon in Austin, and each will be a Model Y with a safety person in the passenger seat. It’s not clear to me at all how that’s any different from giving one of your employees a credit card reader and having them ride along in cars that already exist.

I can’t tell what argument you’re trying to form here, so I’ll just reply with a few related things.

Clearly, the big deal here is that Tesla now has a Level 4 system. There is no one in the driver seat and nothing that requires a passenger to ever take over (this is distinguished from lesser versions of remote control to get the car unstuck or otherwise, which Waymo also has). There is a stop button, but it’s just to indicate that the passenger wants to get out now.

The Tesla employee is more there for passenger safety than anything driving related. And, frankly, to ensure no one does anything stupid to get on the news (packing as many people into the back seat as possible, that kind of thing). They’ve been asking passengers to show them the app on their phone. Otherwise they’ve just been giving a short spiel and doing nothing otherwise.

There’s no need for a credit card reader since their app and backend infrastructure appear to be working smoothly. This sort of thing is a much easier problem than FSD but still has to be done. So far, it looks like the payment integration, route searching, etc. are all working well. They’re currently charging a flat rate of $4.20 for rides. There is a joke in the app where you can tip the driver, but it says “just kidding” when you try to submit it.

In a sense, you’re right that this is no different than what Teslas can currently do, aside from some minor software differences and the fact that Tesla is now legally allowed to operate a driverless fleet. That’s the whole point. The cars are identical to standard Model Ys coming out of the factory. This has been their goal from the beginning, and a major factor in eschewing LIDAR. The approach seems to have paid off.

I do wonder if we’ll eventually see Model Y’s without a steering wheel. Maybe they could combine the parts for LHD and RHD cars for a seamless dash (that still has the necessary airbags, etc.). They’ll need that if they want passengers in the front-left seat.

Right now the service area is about half that of Waymo’s. I’m sure that’ll expand over time, but it’s not a bad start. The rides I’ve seen so far look just about perfect, with some minor glitches in the pickup point (which can also be a problem with Waymo, and Uber for that matter). Of course it helps that the weather is good.

If they have teleoperators (they do) or if that Tesla employee ever finds themselves in the drivers seat for any reason, it isn’t really Level 4, is it?

Edit after more digging: Apparently Level 4 can allow for human intervention in unusual situations, so unless the teleoperators or employee-passenger has to get involved more than expected, I’m going to assume that it meets Level 4.

Yes. L4 can’t require the driver to take over–it has to fail safely. But because the car can only drive under limited conditions, it can sometimes find itself unable to proceed. So a degree of remote intervention is expected.

Again, this is how Waymos already operate. Apparently they can be “driven” remotely but very slowly. I expect they still have much of their safety systems still engaged so that it’s hard to run someone over, say. But it’s enough to, for example, get out of a weird parking lot that the car can’t escape on its own.

I don’t know if the safety monitor is ever expected to move to the driver seat for this sort of thing. But even if so, it would still be L4, since it isn’t a real-time taking over. At any rate, I don’t expect them to last more than a few weeks.

Their current mission control dashboard:
Imgur

Got some video feeds, a weather map, a fleet map, and a few obvious metrics. Nothing too surprising. The vehicle list is too fuzzy to read but I can see some columns that are green or yellow (and red, I’m sure). I’d suppose they have some system that highlights cars that have gotten stuck for too long or have some other alert (like if they’ve gotten into an accident).

Waymo has repeatedly said that their engineers cannot remotely pilot their cars, they can only provide clarifying inputs when the car is unclear about what it’s seeing. This may have changed but I can’t find anything saying it has.

It’s so hard to weight this in terms of an achievement. Yes, the area is about half of what Waymo is operating in by square mileage, but it excludes UT and downtown Austin and so is probably a much lower fraction of difficulty. Yes, the employee is riding shotgun, so theoretically they can reach over and grab the wheel if the car does a phantom turning maneuver. But they promise that they won’t, and therefore it’s L4. I’ve read they don’t intend on operating in the rain at this stage, but it barely rains in Austin.

I guess it’s something? But it seems way too premature to start talking about removing the steering wheel. Honestly, this reeks of “Our stock is tanking, we need to get something out the door.” I’m sure they will expand, and I’m sure it will improve, but Tesla has earned nobody’s trust on their timelines, so whether that’s weeks or years (or still decades) is anyone’s guess.

Volkswagen is going to offer a robotaxi version of the ID Buzz, its electric version of the VW Microbus. It’s scheduled to be available as a public transit option in Hamburg. “The ID Buzz AD is the first vehicle in Germany to reach SAE International’s threshold for Level 4 autonomous driving, meaning that the car can drive itself, with no need for a driver behind the wheel, within designated areas.”

Elsewhere, I read that they’re going to make the robotaxi service available in Los Angeles through Uber.

Chuck Cook, during one of the Robotaxi drives, mentioned that he was in a Waymo that got stuck in an area with a large number of pedestrians, and that Waymo service slowly drove him out of the situation.

This is probably just a matter of interpretation. Obviously they can’t just drive it at full speed remotely by video feed. I expect though that they have a mode where they can direct it to go forward/back/left/right slowly, but with most of the safety systems still enabled.

It’s a major step. Just to quote your OP again:

It took about 8 years, not “decades.” Sure, it’s still geofenced and all that. But there’s no driver and people are riding around with no interventions.

And sure, it’ll be interesting to see how things develop over the coming months. Will there be any major incidents? Will they rapidly expand the service area, both in Austin and other cities? When will they improve the weather tolerance? I don’t know, but this is nevertheless a huge milestone.

Yeah, I really should have accounted for geofenced taxis in my OP. I think that’s fairly far from the future we were all expecting, where we buy a car that just drives us around. And that still feels like a distant goal.

Plus, I don’t live in Austin so I’m over here gettin’ nothing.

Well, now you’re talking about the difference between L4 and L5. That should probably have much finer gradations, since strictly speaking, very few humans drive at L5. I wouldn’t drive in a blizzard. I don’t do serious off-roading. I’m willing to drive 100 mph but not 250 km/h on the Autobahn. Etc.

On the other hand, Tesla currently has basically the bare minimum L4 system, which just works in one part of one city in optimal weather. Waymo has slightly greater bounds, though it still can’t drive between cities or on the highway at all, so we might call it L4.1.

My opinion is that Tesla will scale faster than Waymo, but it will still take at least a year for them to catch up. And it’ll still be L4.2 or whatever we want to call it. It’ll take a few more years before it gets to be the high 4s (say, it works on all public roads in the country). Probably >10 yrs since the thread start but <20 years.

Same deal as Waymo and Tesla:

MOIA also provides the full suite of services necessary to reach Level 4, including the self-driving software stack itself, but also remote operation and assistance if the car does something stupid like, say, get stuck in a drive-through.

Which is fine. That’s where we’re at right now. Still, there’s a big difference between a car capable of L4 vs. actually serving rides. Hopefully they get there soon.

Do the passengers in a driverless Tesla sit in the front or the back? My concern is that there is no obvious physical safety release for the rear doors.

Back. There’s a screen back there for instructions and it must be shown on it. Right now it’s a pilot program and only available to select people all of whom are very familiar with Teslas