> Musk... said during a recent earnings call ...
"My personal guess is that we’ll achieve Full Self-Driving this year at a safety level significantly greater than a person..."
Does this guy ever get tired of saying "this year" and "next year" every year and being wrong every single time? Do his investors?
Except the part where you don't have people throwing _very_ large quantities of money at you based on these estimates, such that when/if you're wrong it could potentially spell disaster for an entire company costing shareholder money and peoples' livelihoods/jobs.
I'm not trying to claim that each individual claim of Musk's has this potential behind it, but equating it to day-to-day situations of underestimating the difficulty of a small task at your day-job oversimplifies the matter way too much and just makes excuses for someone who should TRULY know better.
> at a safety level significantly greater than a person
Have they actually achieved that yet? It seems like the Teslas can only do their Autopilot magic on more-or-less straight motorways with every other car travelling at the same speed in clear weather conditions.
These are circumstances in which human drivers don't have accidents, either.
Have you ever driven one? I got to drive one for a few days.
It handled heavy rain. It knew where the lines were even though I couldn’t even see them myself. It does really well even when there are no lines.
It does turns and exits. It slows down automatically for curves. It even swerved out of the way when someone was merging into me. It was fine with drivers going faster and slower than myself.
Frankly, I have no idea what you’re talking about, especially regarding a clear day and straight lines.
It drove better than me. In my use case, the 2021 Model 3 is already practically self-driving. I legitimately felt less safe while manual operating the vehicle.
I have, yes. It was a pretty scary experience, to be honest. It doesn't anticipate other drivers' actions, it just reacts - so it was always getting into dangerous situations like having to swerve out of the way of people merging into lanes. It worked okay in the dark, but not dark and rain, and not really in the rain particularly, which makes it not very useful on the West Coast of Scotland. It only works on motorways and dual carriageways, which, again, make it not very useful.
It doesn't work on single-lane roads so I wasn't able to find out if it can see deer, sheep, cows, or other animals. It probably can.
It's like being driven by a drunk teenager, because it doesn't anticipate, reacts late, and cannot cope with developing situations.
I don't really see the point in them. Why would you want a self-driving car anyway?
"for legal reasons". After all that is the only reason there is a driver according to tesla.com/autopilot. And that's 5 years old, so surely we should feel comfortable in the back seat by now besides the law
He himself says at least half that time his predictions are wrong. His predictions were always aggressive.
He says what he believes, this topic is where he has totally miscalculated. And he recently admired that but said he just has to look at what he think the current fundamental problem is and when it will be solved.
> Do his investors?
Mhh as an investor I would say, no.
Because for me its not about matching prediction but its about overall strategy and the competition.
If there were millions of self driving taxis on the road from the competition I would suggest that is a big problem. However that is not case.
Massive investment in this is the correct thing to do in my opinion. Their approach is correct in my opinion, they are trying to solve the hard problem. They have done many things that no competitor has even remotely done.
So I am happy even if he miss predicts this for the next 4 years unless there is continues progress and the competition has not 100% solved the problem.
Loving the Bible-themed whataboutism. Musk has definitely paid for some extremely cool stuff to happen, but it's fair to point out that he's been making this particular promise since 2014.
And nobody dies if I break a promise. Autopilot/FSD has claimed quite a few lives, in large part because it doesn't live up to the hype coming out of Musk's mouth.
So you count anybody that fell asleep or didn't pay attention as a 'kill' for Autopilot (and that number is very small) while you totally ignore all the times the same software prevented accidents (a very large number).
I guess that one way to analyses issues if you already have a result that you want to happen.
Based on the CNBC article, here's my understanding. Someone please correct me if I'm wrong, because this sounds... really bad with safety implications that are not at all related to self-driving.
1. Tesla vehicles are drive-by-wire.
2. The missing component is a backup chip in the steering system.
3. The backup was excluded in a time frame that precluded robustness testing.
Are any of these items wrong? Because, if not, characterizing this as "oh well you have to bring your car in if you want our full self-driving magic" seems like a massive PR spin and a huge disservice to Tesla owners who every day bet their lives on Tesla's drive by wire system.
(Missing/not using a backup chip on a safety-critical system is exactly what caused the failure of the flight characteristic modification software on the 787-MAX. Missing backups was also identified as a possible root cause in the Toyota deaths. Backups are important, and the history of safety-critical systems that ignored the need for backups is littered with human bodies. Shit happens at scale.)
Of course it's drive-by-wire, it's an electric car. How could it possibly be drive-by-cable? Cars have been largely drive-by-wire since the early 2000s.
The confusion here is between throttle-by-wire (which, as you point out, is now ubiquitous) and steer-by-wire, which Teslas still aren't. There was some hope that the newest yoke-equipped Model S would be, but my understanding is that it's still a direct mechanical link, i.e. wheels still turn if you move the steering wheel when the car is off. The driver-assist functions use motors to move the steering linkage.
Wouldn't steer-by-wire that reduce steering feedback even more? Compared to my old (somewhat unreliable) Jetta with hydraulic power steering, every electric steering-assist car I've driven has terrible steering feel, like I'm driving an RC car. I feel so disconnected from the road, as if I'm merely aiming the car rather than actually driving it.
Right; the whole conversation re: "what is drive by wire anyways?" is a massive red herring. That's really my bad. The operative question is just "what happens if the primary chip fails?"
If the answer is "nothing bad, yet, but maybe something bad when we ship FSD" then Tesla's approach here seems reasonable (modulo not informing customers, which will probably catch them a fine in at least some of these jurisdictions, but in my book is more like a "mild breach of commercial contract" issue than a "fundamental consumer rights" issue).
If the answer is "something bad" then this is a big deal and could be bad news for Tesla.
It can be drive by cable in a multitude of ways. One way is that your foot moves a plug in and out that physically connects the batteries to the motor. There are other ways that are real, too.
Is that really true? I know that there's a mechanical connection, but the existence of a mechanical connection doesn't necessarily mean that the system is not effective drive by wire (for the purpose of reasoning about system safety).
I guess the operative question is: if the controller fails, what happens? If the answer is "loss of operator control", then saying that the car isn't drive by wire is the worst type of "technically true". Just like it was, in a silly sort of way, technically true that the 787-MAX's flight characteristics weren't changed in a way that required retraining...
What is the safety-critical use case of the steering column controller? Just interrupting the ADAS?
So, the extra chip was only needed in case the user bought the FSD package. In view of the chip shortage, Tesla decided not to install it, allowing them to produce twice as many cars. Of course, for those cumstomers who choose to purchase the $12.000 FSD package, Tesla will install the chip for free. …
Well, they're removing hardware (due to shortages) for a feature that's still a few "this year, I promise!" lies away anyway, so what's the big deal. I'm biased against the dishonest car salesman CEO, but if you want to buy a bucket of lies, how's an extra lie going to hurt.
> Well, they're removing hardware (due to shortages) for a feature that's still a few "this year, I promise!" lies away anyway
Here's why I'm confused: in that case why not just remove both chips?
Having no backup for a component involved in the steering system seems like a bad idea for any level of ADAS, including L1, and possibly even fully manual depending on how the steering system works. At least without a very thorough analysis of system behavior in various failure modes, which... is that possible in 3 weeks?
But then, Boeing convinced the FAA that backups weren't necessary for the 787-MAX sensors because a component that was modifying the flight characteristics could be over-ridden by a human and therefore wasn't safety-critical... so, WCGW, really?
It was my understanding that the second chip and component set was there purely as a backup before they decided they needed the extra compute that was essentially wasted.
Wonder if they only nixed it from cars ordered without FSD at checkout.
Does this guy ever get tired of saying "this year" and "next year" every year and being wrong every single time? Do his investors?