Self Driving Car thread

[sporkman]

I've spent enough time with Tesla's self driving system to be really unimpressed with it. I suspect that the underlying tech is fine, the problem is the regulatory/liability controls. For instance: when it rains, the Tesla system tends to turn itself off. Super-irritating. Also, the driver attention monitoring is really aggressive and it has a high rate of false positives. Furthermore, when you get a strikeout, you don't even have cruise control anymore.

I really wish that there were a way to jailbreak the autopilot system in Tesla to turn off all of the annoying aspects.

 

[sporkman]

It feels really silly but…I just got yet another week long strikeout for Tesla self driving. I really do not like tesla auto steer. It is way too cautious and it seems geared more towards minimizing possible liability for Tesla rather than being useful.

kind of thinking of installing a comma3x in the Tesla. Ridiculous, but Tesla‘s driver monitoring is way too aggressive and it won’t turn on half the time.

 

[sporkman]

It needs to be really aggressive unless it's truly ready for SAE Level 5, and it's not. That annoying driver monitoring and reluctance to engage if everything is not right is saving your life.

Actually, no, I disagree. I have a comma.ai which has significantly less compute and fewer cameras. The comma.ai handles those situations just fine. The Tesla could too— i think hw4 is like the equivalent of a ps5 or something? The comma3x in comparison uses a snapdragon 845 which might be as powerful as a ps2 or something. It handles all of those situations just fine.

 

[sporkman]

New Mark Rober video that puts Lidar vs Optical (Tesla) to the test in some scenarios (First half is using Lidar around Dinsey for reasons)


View: https://youtu.be/IQJL3htsDyQ?si=vKLVDCK1R7XEIzZF

Highly enjoyable video.

So to elaborate more...I have significant experience driving with a Comma3x self driving system. I am very happy with it. What is great about it is that computer vision works differently than human vision, meaning that when I can barely see the road, the comma3x can see the road better. So I've driven with a comma3x in horrible conditions, snow, rain, etc. The comma3x stays on the road better than I could with my own eyes.

The Tesla in contrast turns itself off because water gets on the lens of the camera.

I'm kind of fed up with Tesla's system.

 

[sporkman]

comma ai openpilot has a new release and it looks significant.

https://blog.comma.ai/098release/

Interestingly, according to the devs, openpilot uses 40% of the snapdragon 845 processor's GPU capability.

 

[sporkman]

I think that we should have level 3 systems and that there should be no driver monitoring at all. I don't trust level 5 at all because with no steering wheel there is no manual override which I think all cars should have regardless. I don't think that any system can be truly autonomous hands free unless it is on rails. So the best is an autopilot system with no driver monitoring and a manual override should the system make a mistake.

Cars are our living rooms and you know what often I'm reaching around the car while driving trying to pick something up that fell off a seat and onto the floor of the car. Or I'll be eating a meal while driving. Or yes looking at my phone for whatever reason I have. I think the car staying centered in the lane while I'm doing these things should be standard and it shouldn't quit on me leaving me in a more dangerous position. It should always be on with no restrictions or penalties.

 

[sporkman]

Uh...?

"You're responsible for the safe operation of the vehicle you are driving is not an increase in expectations, nor is it unreasonable - it's a basic expectation across (at minimum, the EU, EEA and UK. I''m not an expert on the USA, but I undeerstand that driving under the influence is a crime there, so probably similar.

That there exist people who abrogate that expectation does not make it unreasonable - it makes those people a social problem to be managed, regardless of whether they are unfit because of fatigue, medical condition, drug use or just being sociopathic to some extent. Differerent countries may spend more or less effort managing that problem, and to greater or lesser effect, but making it easier to be a social problem is rarely a desirable political outcome.

(As I understand the term) level 3 self driving does not change that responsibility - L3 is defined such that the car is capable of all necessary actions to complete a given journey, but the driver must be capable of taking back control on being prompted to do so, with no particular expectation nor guarantee that such a prompt will always provide a minimum safe time for the driver to regain situational awareness.

I just find it asinine that if the car detects that you're not paying attention, for which especially on tesla's system there are many false positives, it will intentionally make a crash more likely by allowing the car to drift out of lane. In my modern driving record I haven't had an at-fault accident for close to 20 years. A level 3 system with no restrictions would be perfect for me.

 

[sporkman]

raises eyebrow

That is a) a disturbingly weak answer against b) a disturbingly weak metric.

The better questions, which you are probably too biased to answer (this is not particularly a condemnation of you personally):
1. How many times in the last year have you gone into microsleep while driving?
2. How many times in the last year have you done other tasks while driving (eating, texting, gotten in a deep conversation with a passenger or third party, etc.)
3. How many times in the last year have you been in extremis while driving (i.e. where immediate drastic action was needed to prevent an accident)

Here’s what we know from literally decades of human factors research, accidents, and near misses.

1. Humans get bored really easily.
2. Bored humans get distracted or nap.
3. Humans take for bloody ever to get up to speed (reaction time while paying attention is ~0.25 s, while not it’s 2-15 seconds). It gets worse when humans are tired.
4: Humans are really, really bad about normalizing deviance. (As in, based on familiarity we push the envelope and erode safety margins)
5: Humans believe they are much more capable than testing shows.

So while you claim to be, and (giving the unverified benefit of doubt) might be a 6 sigma safest driver, the 200 humans next to you also claim the same thing erroneously. Thus, the human factors engineers tell the systems engineers that the driver alert system is safety critical, and not something for end users to “adjust”.

There have been times when I have been driving while extremely fatigued and going into microsleeps.

My comma.ai system kept me safe going in the lane perfectly. It was a lifesaver. If I didn't have the system it would undoubtedly have been much less safe. Because I experienced such a circumstance in my previous car which wasn't compatible. That circumstance being fatigued with no self driving system was very dangerous. Driver monitoring would basically disable the system when it is most useful.

The Tesla system in contrast is such bullshit. You can't even look at the car controls or the map on Tesla's own screen without it giving a warning. Tesla designed a car with a central touchscreen. It punishes you for even looking at it or using it.

But uh, the comma system does have driver monitoring though it is way better than Tesla's in that it has far fewer false positives. You can't completely fall asleep while using it but you can adjust the sound system.

 

[sporkman]

Then you shouldn’t have been driving, whether you were getting an assist from the car or not.

If you have some extreme circumstance where you literally have no choice (driving towards emergency care from somewhere without cell service or something) then that’s one thing. But since you said “times”, meaning this has happened to you more than once, and taken with the rest of your recent posts in this thread, I’m just coming to the conclusion that you’re a terrible driver. And possibly just mad that the Tesla won’t let you be a terrible driver.

but the thing is that tesla technology can make me a very very good driver in spite of myself. It's fully capable of that, but the company has this overly paranoid driver monitoring system that prevents me and the car from being very very good at driving.

 

[sporkman]

At L3, the car is designed to fail over to you. The manufacturer knows that it can’t handle a significant number of edge cases. The driver monitoring is to ensure that you are ready to take over as the backup system. Driving fatigued means that you are going to incapable of taking over in a timely manner without the monitoring.

hmm, true. that means that what I really need is level 4.

 

[sporkman]

I've never thought that the Uber self driving initiative would be successful because of how the rideshare business actually is. But it appears that a lot of investors would only invest for the promise of self-driving. It's one of those things where the stated objective is never actually achieved but something else can be found out along the way. Kind of like how Slack came out as someone else's internal development tool for another project.

The best use of self driving tech is a better cruise control that lets me eat noodles while driving.

 

[sporkman]

I view the benefit of self driving like a blue light filter on a computer screen. It reduces stress and allows you to be more alert once you get to the destination

 

[sporkman]

Around here (NL/DE) with platooning, the big problem is still just cost. Self-driving trucks are considerably more expensive and have a lot of extra (hidden) labor involved in them, so there's currently no cost or labor savings. It's also not the purpose of these trials and licenses right now - the first step is to gain trust in the system and figure out the kinks. It's a very, very slow process of getting companies to have confidence in the concept, developing systems to incrementally remove all the non-driving labor and end up with a system that does autonomous driving the way it has been promised. A bit like the Rotterdam harbour authority: they took about 20 years to go from on-dock cargo starting to get automated to having reduced >99% of the labor force for - basically - longshoreman-type tasks.

There's also been a lot of discussion academically about the real achievable cost reduction. Drivers aren't that expensive, nor are there that many opportunity costs to using human drivers on most routes. Most arguments around automation don't apply to trucking, e.g. most trucks can't drive when the driver is sleeping... well, yeah most suppliers and customers are also closed at those times. Only trucker shortages are a good argument right now.

I don't think self-driving with no human driver will ever be a thing. At the very least, a human driver also provides security for the cargo and he helps with mechanical breakdowns which an AI cannot do. I don't think an AI could winch down a load.

What AI self driving can do is substantially reduce driver fatigue. This improves safety and allows the driver to work more hours.

Same for rideshare