Scenarios in which Tesla FSD Beta 9.0 fails • Hacker News Board

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#311

Post by Method-X » Mon, Jul 12, 2021, 10:07 PM UTC

CyanLite2 wrote: post #296

Animats wrote: post #203

Earlier quoted context omitted.

The monorail video is jaw-dropping. Yes. Pause the video and look at the car's screen. There's no indication on screen of the columns. A car on the other side of the row of columns is recognized, but not the columns. It's clear that Tesla has a special-purpose recognizer for "car". The columns are a solid obstacle almost impinging into the road, one that doesn't look like a car. That's the standard Tesla fail. Most h…

LIDAR cost is coming down, but Tesla has almost no margins on cars. So adding a $500 part per car for LIDAR is out of the question. Even the cheap radar parts were removed from production going forward. However, it's good to see other automakers like Volvo going the LIDAR route on their 2022 electric models and are offering Level3 autonomous driving.

Tesla’s margin on the Model 3 is 35%:

https://thenextavenue.com/2020/06/10/teslas-model-3-profit-m...

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#312

Post by Invictus0 » Mon, Jul 12, 2021, 10:10 PM UTC

zaroth wrote: post #267

Veserv wrote: post #219

Earlier quoted context omitted.

The rate of driving fatalities is ~1 per 60 million miles driven. At an average speed of 60 mph that would be ~1 per million hours. Even if we were to assume there would be only one fatality-inducing error per hour, 99.99% success rate would make FSD 9 100x more dangerous than the average driver. I do not remember the exact numbers, but that would make FSD something like 1000x more dangerous than the classic example…

We have actual statistics on how well humans do with Tesla ADAS and it’s remarkably well. The accident rate numbers are hard to compare apples-to-apples since miles driven with all ADAS features active are different from miles driven without the full ADAS active, but the numbers show that the software is absolutely not “1000x more dangerous”. > In the 1st quarter [2021], we registered one accident for every 4.19 mill…

> one accident for every 4.19 million miles driven

Not really a great metric. We should be looking at disengagements per MM.

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#313

Post by threeseed » Mon, Jul 12, 2021, 10:11 PM UTC

brianwawok wrote: post #273

BugsJustFindMe wrote: post #262

Earlier quoted context omitted.

The post I was responding to said lidar, not radar. But if you want to switch to radar, we can talk about that too. > camera sees air, radar sees wall, slam on brakes Seeing bridges as walls is not a fundamental property of radar. That's an implementation problem. If cars are doing radar poorly, maybe the fix is to start doing it less poorly instead of throwing it away entirely.

Well, that was specifically about blending two different sensors with different characteristics. For you walking, it would be like blending your eyes with your nose. If your eyes tells you the floor is safe, and your nose smells something bad, do you stop? Anytime you have two different sensors with different characteristics, you want "the best". Your body uses your eyes to see where to walk, and your nose to test if…

> but I think trying to blend the two is a sub-optimal solution to the problem

Research over the last decade has shown that LiDAR/Vision fusion outperforms Vision Only.

Can you explain the science behind your position ?

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#314

Post by raisedbyninjas » Mon, Jul 12, 2021, 10:11 PM UTC

CyanLite2 wrote: post #296

Animats wrote: post #203

Earlier quoted context omitted.

The monorail video is jaw-dropping. Yes. Pause the video and look at the car's screen. There's no indication on screen of the columns. A car on the other side of the row of columns is recognized, but not the columns. It's clear that Tesla has a special-purpose recognizer for "car". The columns are a solid obstacle almost impinging into the road, one that doesn't look like a car. That's the standard Tesla fail. Most h…

LIDAR cost is coming down, but Tesla has almost no margins on cars. So adding a $500 part per car for LIDAR is out of the question. Even the cheap radar parts were removed from production going forward. However, it's good to see other automakers like Volvo going the LIDAR route on their 2022 electric models and are offering Level3 autonomous driving.

They currently charge $10k for FSD. Even if they don't drop any of their cameras and processing equipment, $500 LIDAR is a 5% increase. Is it just me or are we politely ignoring Musk's reluctance to admit his bold proclamations against LIDAR are the biggest factor in their design decisions.

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#315

Post by threeseed » Mon, Jul 12, 2021, 10:13 PM UTC

serverholic wrote: post #72

Syonyk wrote: post #11

I don't understand how something this broken is allowed to operate on public roads. If I drove like Tesla's FSD seems to based on the videos I've seen, I'd be pulled out of my car and arrested on (well founded) suspicions of "driving while hammered." After a decade of work, it's not capable of doing much beyond "blundering around a city mostly without hitting stuff, but pay attention, because it'll try to hit the mos…

We are going to need a fundamental breakthrough in AI to achieve the level of FSD that people expect. Like a convolutional neural network level breakthrough.

Cruise has videos showing them driving around for hours, under challenging conditions with no issues:

https://www.youtube.com/channel/UCP1rvCYiruh4SDHyPqcxlJw

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#316

Post by paxys » Mon, Jul 12, 2021, 10:14 PM UTC

the8472 wrote: post #201

j7ake wrote: post #156

These spectacular fails weaken the rationalist's arguments that "as long as FSD achieves lower deaths per km driven (or any other metric) than humans" then FSD should be accepted in favor of human driving. Even if "on aggregate" FSD performs safer (by some metric) than humans, as long as FSD continues to fail in a way that would have been easily preventable had a human been at the wheel, FSD will not be accepted into…

I think you misunderstand the argument. It is that if, hypothetically, FSD really did save human lives on average then it should be accepted as the default mode of driving. It would be a net win in human lives after all. But the "should" can also acknowledge that people irrationally won't accept this net life-saving technology because it will redistribute the deaths in ways which they're not accustomed to. So it's as…

I don't agree with the former argument either. I'm not going to accept a self driving system unless it increases my personal safety. If the system doubles my accident rate but cuts that of drunks by a factor of 10 (thus improving the national average), it isn't irrational to not want it for myself.

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#317

Post by CyanLite2 » Mon, Jul 12, 2021, 10:15 PM UTC

potatolicious wrote: post #213

bobsomers wrote: post #92

Earlier quoted context omitted.

I don't think that's necessarily true. There are plenty of people in the AV space that routinely drive significantly better than this and are already doing completely driverless service in some geofences. The problem with Tesla's approach has always been that Elon wanted to sell it before he actually knew anything about how to solve it. It's lead to series of compounding errors in Tesla's approach to AVs. The vehicle…

Yep, also over-promising that legacy cars with the "FSD-readiness" upgrade would be guaranteed to support FSD was a completely unforced error. At this point it's pretty obvious none of the original "FSD ready" cars Tesla shipped will be able to handle L4/L5 autonomy hardware-wise. Arguably none of the cars currently rolling off the line will be either - others have made the point in the thread: while nobody has reall…

I think there's a clear space in North America for a viable L3 system. That is, completely hands-free driving at slow speeds (30-50mph) that allows me to eat a sandwich, respond to a few emails, and text on my phone without having to look at the road or grab the wheel every 15 seconds. Sure, ping me if I'm coming up on a construction zone or if it pours down raining. And don't worry about making unprotected lefts and other tricky situations, I can do that. But just keep the car on the road while I'm making the same drive to the office every morning and let me reclaim some of that commute time.

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#318

Post by belter » Mon, Jul 12, 2021, 10:16 PM UTC

bob1029 wrote: post #119

serverholic wrote: post #72

Earlier quoted context omitted.

We are going to need a fundamental breakthrough in AI to achieve the level of FSD that people expect. Like a convolutional neural network level breakthrough.

I think we should first start with a fundamental breakthrough in our willingness as engineers to admit defeat in the face of underestimated complexity. Once we take proper inventory of where we are at, we may find that we are still so far off the mark that we would be inclined to throw away all of our current designs and start over again from (new) first principles. The notion that you can iteratively solve the full-…

In other words, and as somebody said:

"Just because we can put a man on the Moon does not mean we can put a man on the Sun" :-)

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#319

Post by bumby » Mon, Jul 12, 2021, 10:18 PM UTC

baud147258 wrote: post #307

bumby wrote: post #290

Earlier quoted context omitted.

> That is what happens now with phantom braking. Uber seemed to have programmed an artificial delay when the system got confused. There's a good breakdown of the timeline showing how the system kept misclassifying the bicyclist who was killed, but I couldn't immediately find it. That breakdown shows their strategy at implementing delays in the decision process. According to the NTSB report[1]: > "According to Uber, e…

> "we wanted to avoid nuisance braking so we put in a delay when the system was confused." As someone who used to work in safety-critical software, it blows my mind that you would deliberately hobble one of your main risk mitigations because your system gets confused Maybe it was put in place to avoid erratic braking in absence of obstacle, in order to avoid getting hit in the rear by other vehicles (whose driver wou…

That's a good point and might have been their rationale, but I would argue it wasn't a very good risk mitigation because while they reduced the risk in one area (being rear ended) they increased their risk elsewhere. Worse yet, it increased the risk in an area more prone to higher severity incidents (e.g., hitting pedestrians - I assume - carries a higher severity than being rear-ended)

Re: Scenarios in which Tesla FSD Beta 9.0 fails

#320

Post by the8472 » Mon, Jul 12, 2021, 10:18 PM UTC

paxys wrote: post #316

the8472 wrote: post #201

Earlier quoted context omitted.

I think you misunderstand the argument. It is that if, hypothetically, FSD really did save human lives on average then it should be accepted as the default mode of driving. It would be a net win in human lives after all. But the "should" can also acknowledge that people irrationally won't accept this net life-saving technology because it will redistribute the deaths in ways which they're not accustomed to. So it's as…

I don't agree with the former argument either. I'm not going to accept a self driving system unless it increases my personal safety. If the system doubles my accident rate but cuts that of drunks by a factor of 10 (thus improving the national average), it isn't irrational to not want it for myself.

> The car has to be better at driving than me

But you can also be a pedestrian or passenger. Do you not want everyone else to be less likely to kill you?

Also, should you really trust your own estimate of your driving safety?

> McCormick, Walkey and Green (1986) found similar results in their study, asking 178 participants to evaluate their position on eight different dimensions of driving skills (examples include the "dangerous–safe" dimension and the "considerate–inconsiderate" dimension). Only a small minority rated themselves as below the median, and when all eight dimensions were considered together it was found that almost 80% of participants had evaluated themselves as being an above-average driver.[30]