Self-Driving Cars

It's more appropriate to say that Tesla has a Level 2 system (Autopilot) and a Level 4 system (FSD) putting on the hat of a Level 2 as it is developed. When used as a Level 2 driver-assist system, FSD is vastly more feature-rich than the average competitor — heck, it's way more capable than Autopilot, but it still requires supervision. It's a regulatory grey area.

Anyway, these claims that lidar and radar are necessary to achieve autonomy are not authoritative. They're coming from companies that have decided to invest in those technologies. There's an inherent bias/conflict of interest.

"We'll never achieve self-driving without lidar, says the company manufacturing lidar."

Nobody has ever done this before, so pronouncements about the technology necessary to achieve it are inherently suspect.

Anyway, here's your daily dose of Supervised FSD: yielding to an unprotected left-turn in traffic. That's insanely humanlike behavior.

I realize that some people will say that this is inherently unsafe, but the road in question has only two lanes so there's no risk of someone coming up on the right and crashing into the vehicle that's turning blind. Not unless that vehicle is driving illegally on the shoulder, anyway.

I wonder if FSD would do that on a multi-lane road. I hope not.

Edited by Fighteer on Apr 9th 2024 at 2:55:57 PM

But it’s not a regulatory grey area, it’s Tesla refusing to apply for Lv 3 or Lv 4 certification and instead trying to have their cake and eat it to by having the system be legally classified as Lv 2 to protect them from liability while having it operate like a Lv 4.

What baffles me is why Tesla are playing this game, the system does seem to work very well, so why not take the jump? I can manage a few ideas.

• The system isn’t functioning as well as we think and isn’t ready.
• Tesla are afraid that any mistakes will get blown out of proportion for the first mover so they want someone else to eat to bad PR first.
• Tesla leadership are perfectionists for perfectionist sake and refuse to go for Lv 3 or Lv 4 certification until they’ve solved everything.
• Tesla has a deep hatred of the idea of being liable for their own system so are never planning to move beyond Lv 2 unless someone extends the Lv 2 liability shield to other levels.
• Tesla leadership are being contrarian for the sake of it and don’t want to go for Lv 3 or Lv 4 certification because they resent the idea of asking regulators for permission to do something.

Option 2 is the only one where I have any sympathy for Tesla (because fear of the press blowing things out of proportion is well founded) but I don’t feel able to rule out other options, though I’d say option 1 is probably the least likely.

Or you're vastly overthinking it. FSD is not ready for unsupervised use yet, therefore it's not being promoted or regulated that way.

It is a system that is intended, once mature, to operate as Level 4/5, but for now it requires active supervision and feedback. There's nothing complex or nefarious here.

ETA:

• FSD avoids a gravel spill on a highway off-ramp.
• FSD detects and slows for a pothole.

All of these behaviors need to be comprehensive enough and consistent enough to require no human intervention during the overwhelming majority of drives. That's the standard regulators will expect, and if not it's the standard Tesla will be aiming for.

Edited by Fighteer on Apr 9th 2024 at 5:00:46 AM

Except you just called it “a Level 4 system (FSD) putting on the hat of a Level 2”.

Either it’s mature enough to make decisions on its own (as Tesla argued in the tweets you linked last page about how Tesla want it to be able to do manoeuvre without direct driver authorisation for each manoeuvre) and it’s a Lv 4 system or it’s not and people need to stop acting like it is.

Nobody's "acting like it is". We are talking about its potential, nothing more, although it is getting very close to that potential.

Again, it is capable of driving itself fully autonomously in many conditions, but consistency needs to improve for it to gain the necessary regulatory approval. For now, it is classed as Level 2 because that requires active supervision.

You literally linked to Tweets where a Tesla exec was arguing that as of today the system is ready to safely engage and complete manoeuvres such as a lane change.

You also claimed that there is “a regulatory grey zone”, if we were just discussing potential there would be no grey zone, as the system would simply be Lv 2 and once development reaches the next stage would become Lv 4.

It absolutely is not. By Tesla's own admission, it is a level 2 system that they hope to develop into a level 4 system someday. It is not a level 4 system right now, and it may not ever be a level 4 system.

Waymo and Cruise have level 4 systems. They have vehicles operating on public streets without drivers. This is a thing that is currently happening right now, not a thing that they plan to do later. Granted, right now the conditions that allow for level 4 operation are limited enough that they do not want to release it as a commercial product — but "Full Self-Driving" can't even operate as a level 3 system in any circumstances. Tesla is very insistent that it is a level 2 system.

Ah, but Tesla's insistence that it will have a level 4 system without using the technologies it has chosen not to invest in is pure as driven snow and completely free of any sort of bias or ulterior motive, obviously.

So far the only proven level 4 systems use lidar for extremely obvious reasons. It may be possible to eventually achieve level 4 without it. The ball is currently in Tesla's court to demonstrate that lidar-less level 4 autonomous vehicles are 1) possible, and 2) superior to versions that rely on lidar. I am extremely skeptical due to their long and storied history of overpromising and underdelivering. They are welcome to prove me wrong.

Edited by NativeJovian on Apr 10th 2024 at 6:49:24 AM

It's just weird that Tesla insists on only using the cameras. Do they just want to be different from the competition?

Money saving.

I wonder if they've spent more having to spend years determining information lidar gives from cameras alone than the hardware cost at this point. Especially if the cost might have come down by now.

Musk said that Tesla uses cameras because people drive using their eyes, so cars should do the same.

The guy underneath Musk working on it says it’s cutting out the number of components needed and simplifying the end product.

...<facepalm>

This otoh seems like the real reason.

Edited by M84 on Apr 10th 2024 at 6:22:27 PM

Look, I don't have insider information as to why Tesla hasn't submitted FSD for L4 approval. They probably could at this point, but it's not something I can guess at. I feel like there's some disconnect here about intentions vs. reality.

The intent is to deliver an L4/5 capable system. If it is not sufficiently reliable for that yet, then keeping it at L2 allows it to occupy a different regulatory regime that requires active human supervision. I really don't understand how this is hard to grasp.

Supervised FSD requires human supervision. The goal is to remove the need for supervision.

I believe the phrasing you should be using here is "had", since AFAIK both of them have withdrawn their driverless vehicle pilot programs and are now using safety drivers again. Perhaps that has changed since I last saw news about them, or maybe it's that they aren't allowed to take fares. In any event, they always have human engineering drivers supervising their trips remotely.

In this very thread, at least one person argued that this means they aren't really L4/5; they're just wearing the hat for marketing purposes. I read online, although I would have a very hard time finding the article(s), that all of the monitoring and control systems mean it is far more expensive to operate a Waymo or Cruise vehicle than to just put a human driver in a taxi, and both companies are bleeding cash.

Ah, I see we're back to this sort of argument style. Obviously Tesla is biased in favor of its own approach. Nobody has argued otherwise. What you have failed to provide any support for is the idea that lidar and/or radar are necessarily part of any autonomy solution. First, you're asking to prove a negative, but second, since there are no approved L4/5 systems for general use, there is no hard evidence either way.

There are no "proven level 4 systems" in the way you describe. Waymo and Cruise have had major issues in recent years and have both substantially withdrawn their rates of investment, and as noted above, they are putting a fig leaf on their level of autonomy.

All I'm asking is that we not engage in arbitrary skepticism.

Both of these things can be true. Yes, obviously it saves money to remove lidar and radar. For autonomy to succeed, it needs to be affordable. Putting $100K, or even $10K worth of sensors on every car is prohibitive if the goal is to have all vehicles be self-driving.

The question, which nobody has yet answered, is whether a vision-only approach will work. Tesla is putting a big bet on that, and while we don't yet know if it will succeed, it's certainly looking promising.

I would also like to point out that a thing taking longer than originally promised doesn't mean it's a scam or a lie. It just means that it took longer. That's the risk when you're on the bleeding edge.

Tesla's Rohan Patel posted yesterday:

This is a valid point. If we're so worked up about driver monitoring in vehicles with driver-assistance features, we should be even more concerned about driver monitoring in vehicles without them. Why don't regulators require every car sold to have cabin-monitoring cameras and steering-wheel nags?

While we're at it, let's add breathalyzer ignition interlocks, speed governors if seat belts are not worn, and force the scanning of a valid driver's license before the ignition can be turned.

(Yes, I'm being rhetorical here.)

Edited by Fighteer on Apr 10th 2024 at 11:14:28 AM

Because that feature in a conventional vehicle is "you crash", not "you're breaking the law and will crash if something goes wrong". You know, because instant injury or death is its own motivator. No extra incentive needed to be ready to respond.

And never a liability question, there.

Edited by RainehDaze on Apr 10th 2024 at 4:17:10 PM

So, you concede that vehicles with driver-assistance systems are safer than vehicles without given a comparable level of driver attention?

Edit: I love how screwed up our priorities are. If someone rams a bus full of children while looking at their phone, it's just life. Sucks to be the kids. But if someone is looking at their phone while using a driver-assistance system and the same thing happens, all of a sudden we get outraged at the ADAS.

Edited by Fighteer on Apr 10th 2024 at 11:29:36 AM

No, we're still outraged, but someone who gave up control of the vehicle to look at their phone is going to be doing it regardless. Nagging them in an assisted car is to reduce the odds of the people who WOULDN'T do that going "it's fine, car has got it". Instead, car gets to scream at them.

Also crap like people setting the car off without a driver or having sex or even more dumb things that are way trickier to try and pull off whilst controlling it.

No, nobody should be looking at the phone while driving. Yes, there would and should be outrage at the driver of a regular vehicle not paying attention and getting into a bad crash because of it.

The point is, the pitfalls of not paying attention when things go wrong are much more immediate and readily apparent on a normal vehicle. And you're kinda forced to pay at least *some* attention to figure out where to go. But if the car mostly drives itself except if it doesn't properly you have to be immediately paying attention, they want the extra emphasis to make sure the person actually *is* ready to take over. Do you not understand that?

and exactly

Edited by PointMaid on Apr 10th 2024 at 8:30:29 AM

Basically, if you can't rely on perceived risk to make people behave, annoy them or force them into compliance by stopping as soon as safe.

In these situations, the human is at fault either way. That's the point. Blaming the ADAS is missing the forest for the trees. We are hard-wired to look for scapegoats: to find excuses for why the human at the wheel isn't responsible.

Our society has decided to tolerate a certain amount of death and injury on the roads because "what can you do, people gotta get to work". That is the real tragedy here, and one that self-driving is attempting to address.

No, we're saying to not design an ASSISTANCE feature to tacitly support bad behaviour. There's no deep meaning, just risk mitigation.

You COULD try and apply this to all vehicles, but it's easier to restrict new and minority segments.

Edited by RainehDaze on Apr 10th 2024 at 4:37:54 PM

Aha! Yes, this is the obviously true answer. It's easier to pretend we're making things safer by restricting ADAS than by addressing the actual problem: human behavior.

If I turn on my Subaru's traffic-aware cruise control and automatic lane-keeping, then take a nap in the back seat, it's my fault if it crashes into a schoolbus. Full stop, end of story.

Edited by Fighteer on Apr 10th 2024 at 11:40:06 AM

It's safer than not. This persecution complex is ridiculous.

I don't know where you got that idea. I live in Phoenix and I encounter 10-12 driverless Waymos every day during my 15-minute (each way) commute. I've seen precisely one with a safety driver. Not one per day, one in the past year. Now obviously I'm not on every road at all times of the day, but 95% of my commute is on a very busy arterial directly off the 10, so it's pretty representative.

Sorry, I had to go back and refresh my memory. It's Cruise that suspended driverless operations in October, 2023.

Edited by Fighteer on Apr 10th 2024 at 12:39:17 PM