Am I missing something here (EAP expectations)?

[TMeister]

The press and forum traffic are awash with all the news of EAP crashes and who's at fault, etc.

Isn't the fundamental question one of the risk an operator is willing to take? That is, if I have complete trust in the AI driving my car, I'll let it drive and not worry about it anymore. OTOH if I doubt the AI's ability to operate in all circumstances, I'll assess each one and own control when I think I should.

All of the crashes I've read about fall in to first category. People are just too trusting of the tech and taking on too much risk. EAP is nowhere near capable of driving the car anywhere but on interstate highways with well marked lanes. Folks here routinely "test" EAP in situations where I would never allow an automaton to operate. It's just too complicated at present time to be certain that all decisions will be correct. Crests of hills, urban intersections, single lane twisty roads, etc. all are interesting test cases but not places people should be allowing the AI to operate.

EAP was marketed to me as developing tech and nothing more than level 2. The service techs also caution me every time I see them that AP really works best and is designed for the interstate situation. And that's why I bought it. My use case is driving long distances on the interstate where at times I can engage EAP when conditions are mundane. This is where I would get totally bored driving. Those 200-300 miles where I can reduce my steering input, mirror and speed scans keeps me alert for the rest of the day's drive.

Each time I get new FW I do go out and see what EAP can do now and each time I'm not impressed. It still drives well on the interstate, but is not trustworthy anywhere else. It can't reliably determine if a car is or is not in my lane. It cannot determine that a crossing car making a left in front of me will be out of the way by the time I get there. It cannot choose which of the lane markings it should follow when one disappears because of an entrance ramp. It cannot determine that a right turning car will be gone out of my lane when I get there. These situations and more tell me that I cannot rely on EAP to work through these situations so I don't use it.

So I'm wondering why does everyone want EAP to be more than it is?

Yeah I know Elon said so but you should all know that evangelists almost always describe the unattainable future. And by this time given all the disappointments along the way you should not lend a whole lot of credence to what he said or says.

This whole self driving thing reminds me of other tech like fusion that was always 20 years ahead and still remains 20 years ahead 40 years later. Car manufacturers touting level 5 autonomy in the next few years will 10 years from now still be saying it is just a few years away.

 

[daktari]

AP is marketed as increasing safety.
Like a fence on a cliff. But the fence is Styrofoam, and painted as steel.

 

[WannabeOwner]

Isn't the fundamental question one of the risk an operator is willing to take?

A problem I have is that the way that AP mis-behaviour matches absolutely nothing on my expectation list. If I take a teenage for a driving lesson I have a fair idea of the daft things they will do. Suddenly veering out of lane because there is more rain / sun than yesterday just isn't one of those things ... so there is no way for Humans to know how to Trust AP. But the risk is that because AP appears to be very good then people over-trust it (e.g. in the early days drivers made videos of them in the back seat / whatever ...)

Even on a road that AP has successfully driven you many, many times if e.g. the gore-point has been damaged / not reset you are at risk ... where you never were previously.

I drive 99% of my highway, dual-carriageway, miles on AP. I arrive far more refreshed than manual-driving, and because I have read these forums and know about incidents that people have had I am vigilant the whole time, and always have a hand on the wheel. Plenty of people here say hands-on-lap is sufficient, and clearly YouTube is full of people holding a shaky camera in one hand to video the dashboard - about which they give a commentary whilst driving, so clearly part of their concentration is making that video.

So I agree, I think it is very likely that people who have never read these forums (in particular) may be way too overconfident in their use of AP.

always 20 years ahead and still remains 20 years ahead 40 years later.

I don't think that will happen with AP. The AI stuff will progress at exponential rate, so "all of a sudden", to observers, it will be here. Quite possibly this year or next ...

AP is marketed as increasing safety.
Like a fence on a cliff. But the fence is Styrofoam, and painted as steel.

Its not just marketing, the Statistical data proves it. Tesla have data for the MS fleet before AP was activated, and have thus been able to directly compare that with after AP was activated. Insurance in the UK will give you a discount if you have AP fitted (of course they have no idea if you actually use it ...). So IMO its not a Styrofoam fence.

As I've said before in these forums, I'm happy that either AP or ME detects trouble ... that's better than me on my own.

 

[daktari]

Its not just marketing, the Statistical data proves it.

Sorry, but no. I would say there are too many confounders that are not accounted for in Tesla's claims.

 

[RayW]

I wonder if any drivers had the expectation cruise control could handle stop-and-go traffic in the 1950’s, and blamed the technology when they rear-ended the car in front of them?

 

[WannabeOwner]

I would say there are too many confounders that are not accounted for in Tesla's claims.

Well, of course depends whether you believe Tesla, but (for me) not to do so is to assume conspiracy. Not impossible of course ...

AP accidents well publicised ... it makes good Copy for journalists of course. 2 AP fatal accidents (I think?) in a couple of years, both drivers, sadly, not paying attention.

I have not seen many other cases mentioned of AP-accidents (Firetruck. One or two other claimed, which Tesla have dismissed (although hard to say if "AP disengaged just before accident as part of user-intervention, or "Driver was lying/mistaken")

I have no idea what the likely accident rate would be given the [not huge] number of Tesla cars (with AP) on the roads compared to "national average for cars less than a couple of years old", so I don't know if the figure is high, or not. Either way, such a s low number is statistically insignificant, so could be that Tesla have been lucky ... or unlucky ... or could have been far more unlucky but still statistically insignificant.

Is the number of known AP accidents / fatalities, relative to cars-sold, High? low? Impossible to say?

 

[daktari]

Is the number of known AP accidents / fatalities, relative to cars-sold, High? low? Impossible to say?

Probably impossible to say, because number of incidents are so low compared to number of vehicles and kilometers travelled. I found some reports from Norway, but not my field and hard to translate. Anyway; accident frequency seem to range from 0,01 to 1,0 pr million vehicles travelled at a special road type/crossing etc. This makes for really difficult interpretations in my experience.

 

[MarcusMaximus]

Sorry, but no. I would say there are too many confounders that are not accounted for in Tesla's claims.

While true, it does suggest that it *likely* improves safety. There will always be confounds in this kind of real-world data. But that doesn't completely invalidate the real-world data or suggest that the opposite of what it suggests is likely to be true.