Tesla Full Self Driving requires human intervention every 13 miles

That's really why "miles" is a poor measure. It implies going faster is safer.

It would be better to list intervention per hour and then list the categories of driving (city, rural, highway).

>> unless you want to be the biggest nuisance on the road and a total jerk to everyone around you.

That is likely a selling point for many customers. Call it compensation for the lack of loud tailpipes on electric cars.

Highway driving 600 miles is significantly different than driving in a city. You should test it around town first and hope for no regressions. Yes, once through the neighborhood was good, but that's probably also significantly less than 13 miles.

While people certainly appreciate anecdata on self-driving experiences, by the very nature of driving incidents/accidents (very rare events highly unevenly distributed following a power law distribution), anecdata will always be detrimental to the collective understanding of FSD.

We really need to stop sharing personal stories on FSD. What we need is a public statistical description of the situation.

> I was absolutely blown away.

All my acquaintances who have Teslas and FSD were insisting on the same. But every time I hitched a ride as a passenger, most of the times with the explicit goal of seeing this miracle in action, I mainly got a long string of "that shouldn't have happened, usually it's a lot better".

I'll happily take the blame on me and assume I'm a jinx but it's more likely that Tesla owners get really enthusiastic about their purchase and prop it up by committing the good experiences to anyone that will have them, and the "oh God my life flashed before my eyes" moments to /dev/null. Nobody likes the discomfort of post-decision dissonance.

> each release is noticeably better.

FSD beta was introduced four years ago. How many more updates will it take for it to achieve FSD as full self-driving capabilities?

If you sell the car can the software locked features actually be transferred to the next owner? It seems exactly the kind of thing that would be disabled and require the new owner to purchase.

The name being false advertising should be reason enough to never use it.

MTBF of 13 miles, where "failure" could mean "fatal auto crash", sounds pretty terrible to me. That's 5 times per hour at 65 MPH.

I mean during regular driving there are moments requiring intervention that occur at unpredictable times and when you're about to die if not properly responded to, they're just awash in a sea of lower stakes interventions. My commute is about 3 miles and this morning I had two people in front of me slam their brakes unexpectedly and another person swerve into my lane just ahead of me. So that's roughly one high stakes human intervention per mile.

I'm not saying that 13 per mile is good enough, but the metric does seem useful.

There are a lot of smart people on here that are jealous of Musk's accomplishments and/or disagree with his politics.

Running red lights and stop signs is unsafe. I’ve experienced it and there are many videos demonstrating it including very recently.

It’s amazing. It’s better than adaptive cruise control, but it has a long way to go because it’s often enough unsafe.

The problem is that, as it stands today, FSD requires constant human attention to catch its eventual mistakes. Humans are notoriously bad at that, and the promises of FSD make drivers even more oblivious to their surroundings, ultimately worsening this issue.

Like is often the case with these things, it will continue being approached until it's essentially a solved problem.

On the front page right now is the 10-year-old xkcd comic about how determining if a photo is of a bird takes a team of researchers. This is now trivial.

Cars driving themselves are doing better than they were 3 years ago. Three years from now they will be better still. I don't know how long it will take, and I don't know who will get there first, but I will bet you that the people who are currently in their 50s will never have to give up driving, because by the time they are too old to drive, self-driving will be good enough for them to use.

I disagree that a'99%' level won't get good enough for legal liability. Let's say normal human drivers make an error once every 10 miles, but Tesla's FSD makes an error once every 100 miles. The obvious choice is Tesla FSD even though it is not 100% perfect, and in fact it is 10 times better than a human driver. They might even make it illegal to drive without FSD once it becomes widely available. In the end all the speculation about how good it is or can be means absolutely zero because the results will be created in the end by insurance companies or Tesla, which has it's own insurance company. If it actually is causing more crashes, insurance companies will start charging higher rates, less people will buy Tesla's etc. But we aren't seeing that, with FSD already out there, the insurance premiums for Teslas are not skyrocketing to cover all the accidents. I sort of glossed over this but Tesla also provides insurance which guarantees at least one market participant in FSD, if they cannot make money insuring their own cars FSD will fail, if they can make a profitable insurance company insuring them then FSD will succeed.

It's even worse than this: there is no evidence that FSD is safer on highways due to the extremely low number of bad events there.

It would take literally hundreds of millions of miles driven on highways to get statistical significance.

This is also why L3 is going to be much harder than people think.

How do you know when a Waymo vehicle is receiving guidance from remote support?

I took four trips the other day and couldn’t notice any interventions at all. So maybe they are just so sneaky that I can’t notice.

I've taken hundreds of Waymo rides in SF, LA and Phoenix and the car only needed assistance one time when it got stuck in the middle of a protest.

I'm baffled by Tesla's approach to this.

There are two ways cars currently do automatic wipers.

One way uses a very clever, simple, cheap but very effective sensor. Technology Connections did an episode [1] a few months ago on these sensors. The electronics in it are just a couple infrared LEDs and infrared photodiodes.

The other way is to use a camera, such as the camera you already have for your driver assist functions, and try to use computer vision software to figure out how much rain there is.

Pretty much everyone except Tesla uses the first way. Tesla uses the second way.

What baffles me is that of the three options they might have pursued:

1. Use a rain sensor like the ones everyone else uses,

2. Use a camera and computer vision software which doesn't work as well, and

3. Do not have automatic wipers,

it seems to be that #2 is the worst option. Automatic wipers aren't really needed. Once you get used to where the controls are and how they work on a car without automatic wipers adjusting your wipers for the current conditions becomes one of those things you do completely unconsciously.

So if my next car, like my current car, does not have automatic wipers it is not going to bother me.

But if it has them and the don't work well, that very likely will annoy me, at least until I get tired of having to manually intervene and turn auto mode off or until manual intervention becomes something that I do unconsciously and so forget about them. In either case I end up back in a situation equivalent to a car that doesn't have automatic wipers, but with a lower opinion of my car maker.

Thus this seems like a feature that the car company is better off not doing unless theirs will be as good as what everyone else has which currently means using the IR LED/photodiode sensor.

[1] https://www.youtube.com/watch?v=TLm7Q92xMjQ

It's pretty simple --

If you think Tesla is leading and will solve the FSD problem first, the 1T market cap makes sense, and is possibly undervalued. If you hate Elon Musk because you disagree with his politics and dont drive a Tesla w/ FSD, you make comments like this on HN.