Disruption at its finest :)
Just too much real world data.
(i.e. scaled paid service, no drivers, multiple cities, for 1 year+)
You’re smart Darren, and so are other people, you should assume I knew the cars have remote backup operators. Again, you’re smart, you also know why that doesn’t mitigate having a scaled robotaxi service vs. nothing
I doubt you’ll chill out but here’s a little behind the scenes peek for you that also directly address what you’re saying: a big turning point for me was getting a job at Google and realizing Elon was talking big game but there’s 100,000 things you gotta do to run a robot taxi service and Tesla was doing none of them. The epiphany came when I saw the windshield wipers for cameras and lidar.
You might note even a platonically ideal robotaxi service would always have capacity for remote operation.
My replies are at the same level as that which I respond to, never aggressive IMO.
And if you "knew" something about the relevant topic and leave it out, that in itself is part of the dishonesty.
So once you got a job at Google then you felt Waymo was better, hmmm.
Tesla has a robot taxi service that in some cases has nobody in the car. Also everyone that owns a Tesla has experienced FSD in which it goes from A to B without being touched which is the same as it driving by itself. A person just went cross country and back with this. So to say Tesla is doing none of the 100,000 things you think are required, I think that says more about what someone at Google thinks is needed vs what is happening on the ground.
I am not against remote operation in some cases, but those suggesting Waymo has solved this need to admit that it relies heavily on them for basic decision, like what to do when the power goes out at intersections.
No, some edge case that made the cars fail safely during a power outage doesn't compare. If that's the best you can come up with, you've got nothing.
Robotaxi is a separate product. They are fantastic at driving but until they remove supervisors it’s a moot comparison
And your stats comparing to waymo are made up and debunked in the very reddit thread they came from
So Tesla is in a weird state right now. Tesla's highway assist is shit, it's worse than Mercedes previous generation assist after Tesla switched to the end-to-end neural networks. The new MB.Drive Assist Pro is apparently even better.
FSD attempts to work in cities. But it's ridiculously bad, it's worse than useless even in simple city conditions. If I try to turn it on, it attempts to kill me at least once on my route from my office to my home. So other car makers quite sensibly avoided it, until they perfected the technology.
That's why Tesla's stats are BS. "All drivers, all conditions, all vehicles, all roads" versus "Where FSD is even functional".
https://m.youtube.com/watch?v=2MTmSgVYTTQ
https://m.youtube.com/results?search_query=Fsd+14+blizzard
Any more questions?
In cities, it's just shit. If you're using it without paying attention, your driving license has to be revoked and you should never be allowed to drive.
Tesla FSD gives up with the red-hands-of-death panic at this spot: https://maps.app.goo.gl/Cfe9LBzaCLpGSAr99 (edit: fixed the location)
It also misinterprets this signal: https://maps.app.goo.gl/fhZsQtN5LKy59Mpv6 It doesn't have enough resolution to resolve the red left arrow, especially when it's even mildly rainy.
At this intersection, it just gets confused and I have to take over to finish the turn: https://maps.app.goo.gl/DHeBmwpe3pfD6AXc6
You're welcome to try these locations.
> Red hands of death would be sunglare due to your windshield not being clean. I haven't had red hands since 14 came out.
My windshield is completely normal. Not unusually dirty or anything. It's also Seattle. What is the "sunglare"?
These are influencers who have a stake in Tesla. The general consensus from the regular users is that it is really good starting at FSD 14. It's the first version that finally feels complete. I have 5000 miles on FSD 14 with no disengagements. 99% of my driving is FSD. I couldn't say that for any other version. Even my wife has 85% of her driving on FSD and she hated it before. She just tends to drive herself on short drives and in parkings lots, where as I don't. So your take just doens't line up with what people are saying in social media and my personal experience.
> My windshield is completely normal
If it's never been cleaned from the inside, it's a good chance it's not. The off-gassing from new cars causes fog on the inside of the windshield in front of the camera. It might behave ok (or wierd) but when sun hits it you get red hands of death.
You need to clean it yourself or have Tesla do it. They offer it for free. I did mine following this video and it wasn't bad if you have the right tool. After I did this things were completely fine in low direct sun.
I've seen it on multiple forums. Just like a broken record.
> If it's never been cleaned from the inside, it's a good chance it's not.
The camera is clean. I can see that on the dashcam records. And if the system is so fragile that a bit of dust kills it, then it's not good.
The issue with the red-hands-of-death is caused by the forward collision warning, the road there curves and slopes up, so the car gets confused and interprets the car in front as if it's on a collision course. This happens even during manual driving, btw. False FCWs are a common problem, if you check forums, and people are annoyed because it affects their safety score used for Tesla Insurance.
FSD got better than it was 4 years ago. But it's still _nowhere_ near Waymo. You absolutely can not just sit back and snooze while it's driving, you constantly have to be on guard.
You won't see it unless you shine light into it.
> And if the system is so fragile that a bit of dust kills it, then it's not good.
It's not dust, it's fog on the inside of the windshield from offgassing.
> The issue with the red-hands-of-death is caused by the forward collision warning, the road there curves and slopes up, so the car gets confused and interprets the car in front as if it's on a collision course
Of fair enough. I've never seen this, and I used FSD (14) all through the Appalachian mountains.
> FSD got better than it was 4 years ago. But it's still _nowhere_ near Waymo
Fair enough, but FSD is still years ahead of any other system you can buy as a consumer.
I used the latest FSD and Waymo in December. FSD still needs to be supervised. It’s impressive and better than what my Subaru’s lane-keeping software can do. But I can confidently nap in a Waymo. These are totally different products and technology stacks.
The most recent one is: https://media.mbusa.com/releases/release-4889b1d1c66cddc7120...
I'm not talking about some Tesla style last second bullshit where you're supposed to compensate for the deficiencies of the system that supposedly can do the full journey. I mean a route like L2->L3->L2 where L2 is human supervised autonomous driving and L3 is autonomous driving with zero intervention. You can't tell people they're allowed to drink a coffee and then one minute later tell them to supervise the driving.
Interesting because that's just not my experience at all and a lot of other users.
This isn't even close to being right.
If you're just getting me mixed up with another poster, I got my stats from an electrek article supplemented by Waymo's releases: https://waymo.com/safety/impact/
Tesla's tech is also marketed as a full self driving autopilot, not just basic driver assistance like adaptive cruise control.
That's how they're doing the autonomous robotaxis and the cross country drives without anyone touching the steering wheel.
They might get better but how is that not evidence enough that currently Robotaxis are behind Waymos in self driving capabilities?
FSD is here, it wasn't 3 or 4 years ago when I first bought a Tesla, but today it's incredible.
For better or worse, passive optical is much more robust against these types of risks. This doesn't matter much when LIDAR is relatively rare but that can't be assumed to remain the case forever.
What's crazy to me is that anyone would think that anything short of ASI could take image based world understanding to true FSD. Tesla tried to replicate human response, ~"because humans only have eyes" but largely without even stereoscopic vision, ffs.
Sure, someone can put up a wall painted to look like a road, but we have about a century of experience that people will generally not do that. And if they do it's easy to understand why that was an issue, and both fixing the issue (removing the mural) and punishing any malicious attempt at doing this would be swift
Is this a joke? Graffiti is now punishable and enforced by whom exactly? Who decides what constitutes an illegal image? How do you catch them? What if vision-only FSD sees a city-sanctioned brick building's mural as an actual sunset?
So you agree that all we need is AGI and human-equal sensors for Tesla-style FSD, but wait... plus some "swift" enforcement force for illegal murals? I love this, I have had heath issues recently, and I have not laughed this hard for a while. Thank you.
Hell, at the last "Tesla AI Day," Musk himself said ~"FSD basically requires AGI" - so he is well aware.
But what if your city hired you to paint a sunset mural on a wall, and then a vision-only system killed a family of four by driving into it, during some "edge case" lighting situation?
I would like to think that we would apply "security is an onion" to our physical safety as well. Stereo vision + lidar + radar + ultrasonic? Would that not be the least that we could do as technologists?
LIDAR has much more in common with ordinary radar (it is in the name, after all) and is similarly susceptible to interference.
Like GPS, LIDAR can be jammed or spoofed by intentional actors, of course. That part's not so easy to hand-wave away, but someone who wants to screw with road traffic will certainly have easier ways to do it.
For rotating pulsed lidar, this really isn't the case. It's possible, but certainly not trivial. The challenge is that eye safety is determined by the energy in a pulse, but detection range is determined by the power of a pulse, driving towards minimum pulse width for a given lens size. This width is under 10 ns, and leaning closer to 2-4 ns for more modern systems. With laser diode currents in the tens of amps range, producing a gaussian pulse this width is already a challenging inductance-minimization problem -- think GaN, thin PCBs, wire-bonded LDs etc to get loop area down. And an inductance-limited pulse is inherently gaussian. To play any anti-interference games means being able to modulate the pulse more finely than that, without increasing the effective pulse width enough to make you uncompetitive on range. This is hard.
Large numbers of bits per unit of time are what it takes to make two sequences correlate (or not), and large numbers of bits per unit of time are not a problem in this business. Signal power limits imposed by eye safety requirements will kick in long after noise limits imposed by Shannon-Hartley.
I haven't seen a system that does anti-interference across multiple pulses, as opposed to by shaping individual pulses. (I've seen systems that introduce random jitter across multiple pulses to de-correlate interference, but that's a bit different.) The issue is you really do get a hell of a lot of data out of a single pulse, and for interesting objects (thin poles, power lines) there's not a lot of correlation between adjacent pulses -- you can't always assume properties across multiple pulses without having to throw away data from single data-carrying pulses.
Edit: Another way of saying this -- your revisit rate to a specific point of interference is around 20 Hz. That's just not a lot of bits per unit time.
> Signal power limits imposed by eye safety requirements will kick in long after noise limits imposed by Shannon-Hartley.
I can believe this is true for FMCW lidar, but I know it to be untrue for pulsed lidar. Perhaps we're discussing different systems?
My naive assumption would be that they would do exactly that. In fact, offhand, I don't know how else I'd go about it. When emitting pulses every X ns, I might envision using a long LFSR whose low-order bit specifies whether to skip the next X-ns time slot or not. Every car gets its own lidar seed, just like it gets its own key fob seed now.
Then, when listening for returned pulses, the receiver would correlate against the same sequence. Echoes from fixed objects would be represented by a constant lag, while those from moving ones would be "Doppler-shifted" in time and show up at varying lags.
So yes, you'd lose some energy due to dead time that you'd otherwise fill with a constant pulse train, but the processing gain from the correlator would presumably make up for that and then some. Why wouldn't existing systems do something like this?
I've never designed a lidar, but I can't believe there's anything to the multiple-access problem that wasn't already well-known in the 1970s. What else needs to be invented, other than implementation and integration details?
Edit re: the 20 Hz constraint, that's one area where our assumptions probably diverge. The output might be 20 Hz but internally, why wouldn't you be working with millions of individual pulses per frame? Lasers are freaking fast and so are photodiodes, given synchronous detection.
A typical long range rotating pulsed lidar rotates at ~20 Hz, has 32 - 64 vertical channels (with spacing not necessarily uniform), and fires each channel's laser at around 20 kHz. This gives vertical channel spacing on the order of 1°, and horizontal channel spacing on the order of 0.3°. The perception folks assure me that having horizontal data orders of magnitude denser than vertical data doesn't really add value to them; and going to a higher pulse rate runs into the issue of self-interference between channels, which is much more annoying to deal with then interference from other lidars.
If you want to take that 20 kHz to 200 kHz, you first run into the fact that there can now be 10 pulses in flight at the same time... and that you're trying to detect low-photon-count events with an APD or SPAD outputting nanoamps within a few inches of a laser driver putting generating nanosecond pulses at tens of amps. That's a lot of additional noise! And even then, you have an 0.03° spacing between pulses, which means that successive pulses don't even overlap at max range with a typical spot diameter of 1" - 2" -- so depending on the surfaces you're hitting, on their continuity as seen by you, you still can't really say anything about the expected time alignment of adjacent pulses. Taking this to 2 MHz would let you guarantee some overlap for a handful of pulses, but only some... and that's still not a lot of samples to correlate. And of course your laser power usage and thermal challenges just went up two orders of magnitude...