-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
diplomat33
Average guy who loves autonomous vehicles
Correct. So why are you using an example from April to try to suggest that Waymo is being misleading in their ODD? Maybe Waymo can now handle all fog that humans can handle and therefore it is perfectly appropriate to list fog as part of the ODD.
willow_hiller
Well-Known Member
It's the last most recent example we have of Waymo entering the MRC in dense fog. I'm reading San Francisco's fog season starts in April, so we'll probably need to wait 3 more months until their improvements can be tested in the real world.
I've found 2 ride-along videos in fog published since April last year, and the first isn't very informative because the fog clears up about half-way into the ride:
Also it's difficult to quantify how foggy these videos are. The second fogs up the windows pretty well, but visibility doesn't appear to be impacted too badly, as you can still easily see traffic lights from 1-2 blocks away.
diplomat33
Average guy who loves autonomous vehicles
Daniel in SD
(supervised)
willow_hiller
Well-Known Member
It's conclusive that Waymo can handle some fog. The density of fog that they can handle is unclear.
I've tried looking up fog density metrics, to see if we could evaluate whether it's been foggier than April 11, 2023 since Waymo updated their reliability in fog; but as far as I can tell, no meteorological organization measures fog density or provides historical data on it.
But given that Waymo listed all inclement weather, and explicitly fog, within their ODD, I would expect them to be able to reliably handle all types of weather that a human driver could reasonably handle.
Because I don't think Waymo would consider temporarily entering a MRC as a disengagement, the only way to prove or disprove whether they can actually reliably handle inclement weather is to wait for the convergence of:
1. Inclement weather that humans can still reasonably drive in
2. Waymo vehicles ceasing the driving task as a result of that inclement weather
3. Someone observing it and reporting it somewhere
diplomat33
Average guy who loves autonomous vehicles
Yes, that is my assumption. That is what ODD implies. I don't think they would list fog in their ODD if this was not the case.
Appreciate the thought, thanks. To be clear, I'm saying hat I have no choice but to bet on Tesla at this time - there is nothing else to buy or wait for in the short term.
I'm perfectly open to Mobileye Chauffeur L4, would have been open to GM Ultra Cruise had it progressed, and would open to some Chinese self-driving system. I'm saying that only Tesla has something now; it assists in driving everywhere but cannot be legally or actually depended upon, in its current state.
So I follow the progress of all these systems at some level, and I think I have a fairly realistic view. I kind of wanted to share for folks who read the TMC forums along with eager media FUD. They might easiily come away with the impression that FSD is laughably bad, and that Tesla's come-uppance is right around the corner. but IMO this is an internet distortion.
diplomat33
Average guy who loves autonomous vehicles
If you don't mind me asking, how many more years do you have left where you can drive? You don't need to answer if it too personal. I only ask because if you have say 3 or more years left of good vision, then Tesla is not your only bet. In 3+ years, you will have plenty of "L2+" options better than Tesla FSD. And Mobileye Chauffeur which is eyes-off is likely coming to the US in 3 years. So if you are able to wait that long, that will likely be an option for you.
But if you only have say 1 year left, nobody is your bet because nobody, not Tesla, is going to give you eyes-off on a consumer car in 1 year. IMO FSD beta will likely still require you to supervise it for at least 2 more years. And that is probably crazy optimistic.
Last edited:
Once one group of humans proves that something is possible by achieving it, other groups always succeed in doing the same. An example is the atomic bomb. Many in US thought it to be so technically advanced it would be a US monopoly for decades. Reality proved different. Even back then, I doubt anyone thought no other country would ever have the A-bomb, just that it wouldn't happen in their lifetime.
IMO, if Tesla solves truly autonomous FSD, then others will follow. It is only a matter of time.
GSP
diplomat33
Average guy who loves autonomous vehicles
Even if Tesla does not solve autonomous driving first, others will eventually solve it. I believe it is only a matter of time until autonomous driving is common place. The nature of tech is that someone is usually first but the tech is "rough" and then others come and refine and perfect the tech. Autonomous driving will be the same. I firmly believe that in 5 years, we will see multiple companies with eyes-off or driverless systems.
I am in a similar situation as @JHCCAZ and in my case I don't really know when I will not be able to drive. Likewise, I don't see anything else that I can buy today, so Tesla it is. When similar products are available to buy in the US, then I will consider them. For now, it is just speculation as to when that will really happen.
If my vision no longer is good enough to drive in one year (not likely, but no one knows the future), there are plenty of reasons that I prefer to drive my Teslas until then. They are great to drive and own, better than any other cars, AFAIK. I purchased FSD so that I can experience first hand the "tip of the spear" in research to achieve autonomous driving, and see first hand how it is progressing.
GSP
scottf200
Well-Known Member
GM Is Not Cancelling Ultra Cruise, Just Merging It With Super Cruise
GM's decision to adjust its automated driving assistance features is more a question of marketing than anything.
insideevs.com
The reason, a GM official said, kind of comes down to marketing: Ultra Cruise is about to be merged with the much more well-known Super Cruise.
"GM is not scaling back its advanced driver assistance (ADAS) programs," GM spokesperson Aimee Ridella told InsideEVs today. "We have reallocated our ADAS-focused resources to bring even more capability to Super Cruise under one recognizable consumer brand."
This comes after reports that indicated the Ultra Cruise program was to be canceled, which is only partially true. The key part of the original CNBC story this came from, sources said, is this: "GM has decided to instead focus on the current Super Cruise system and expanding its capabilities rather than having two different, similarly named systems."
<<<<<<<<<<
diplomat33
Average guy who loves autonomous vehicles
>>>>>>>>>>>GM Is Not Cancelling Ultra Cruise, Just Merging It With Super Cruise
GM's decision to adjust its automated driving assistance features is more a question of marketing than anything.
The reason, a GM official said, kind of comes down to marketing: Ultra Cruise is about to be merged with the much more well-known Super Cruise.
"GM is not scaling back its advanced driver assistance (ADAS) programs," GM spokesperson Aimee Ridella told InsideEVs today. "We have reallocated our ADAS-focused resources to bring even more capability to Super Cruise under one recognizable consumer brand."
This comes after reports that indicated the Ultra Cruise program was to be canceled, which is only partially true. The key part of the original CNBC story this came from, sources said, is this: "GM has decided to instead focus on the current Super Cruise system and expanding its capabilities rather than having two different, similarly named systems."
<<<<<<<<<<
Thanks. That is helpful and it makes sense. So that answers my question earlier: GM plans to expand Super Cruise to eventually include what Ultra Cruise was going to be.
This is impossible to say. Without making it a long story, it's progressive vision decline but not entirely predictable. At this point, I can drive anything but I far prefer to drive the Tesla on FSD beta. I don't trust it enough for UPLs in high speed traffic, so I try to avoid those on my route and when needed I control them myself. UPRs to enter the road are more of a Trust but Verify thing. School zones and school buses, I do myself. Everything else is pretty much handled well by FSD in terms of safety, but with the lane selection and turn signal annoyances we've all talked about.
Do I think L4 can/will come to existing Teslas soon enough for me, or even at all? I give those about one out of three and then two out of three, respectively. I'm then more optimistic that L4 will come to a redesigned " Robotaxi" Tesla platform around 2026 and I really want it to be available and affordable for personal nonfleet use. I believe Mobileye Chauffeur on Polestar is looking like the most likely purchasable alternative around 2026, but we'll have to see over the next couple of years.
Despite confident pronouncements here on TMC that Tesla will "never take liability", I think it's quite obvious that the whole Tesla insurance venture was created not for general vertical integration purposes*, but rather as a Tesla-controlled enabler for self-driving liability, data collection and logistical support, so as not to depend on the whims of traditional auto Insurance providers.
*Not to dismiss the profit opportunity from Tesla drivers in the meantime. BTW I personally use State Farm but would be happy to switch for L4 coverage.
And the fact that this growing, consumer-facibg insurance support infrastructure exists within Tesla is one of the better indications that they do intend to support personal autonomous cars. The biggest countervailing argument is that they may well deprioritize and delay the personal-use customer segment as long as high-duty-cycle, fare-earning robotaxi demand exceeds vehicle production supply (the likely initial scenario unless Tesla L4 autonomy really does come to existing cars, something that I just gave only a two out of three to happen eventually - many others are sure it's a zero chance ever).
So while I agree with you that there will be more options, I think there will be very few within 3 to 4 years. It's not a fanboy thing, though I am most certainly a fan of my opportunity to experience and contribute to FSD for now. I would jump for the first remotely affordable solution, but right now Tesla's the only game in town.
That statement is still extremely vague and can mean a number of things
1) There are no change in actual plans (meaning Super Cruise remains as Mobileye based, and Ultra Cruise using Qualcomm) and the only change is in the marketing name (literally renaming Ultra Cruise to Super Cruise)
2) Both Super Cruise and Ultra Cruise will use Mobileye and GM is abandoning the Qualcomm solution
3) Super Cruise will abandon Mobileye and eventually switch over to Qualcomm
Yes, sometimes it happens that a first generation implementation of new technology is "rough". But there is actually another pattern that is quite common: the first generation is good or even excellent, but to get there it's over-engineered and expensive. It's classic that the second generation of a new product is often the real money maker. It takes time to simplify and learn what corners can be cut (cutsthat would have properly made people nervous in the first generation), but also to refine the specification as the users and marketplace learns what's really important.
After all this becomes really clear, it's then a third generation or later that whittles the cost down further, by highly customized and high volume solutions. Single chip Integrations of what was originally a board full of electronics. A drive motor stripped of its original general-purpose housing and bearing mounts, now cast into the product chassis. Sensors and transducers similarly simplified and integrated into a purpose-built assembly.
One of the admirable and unusual things we see from Tesla (and SpaceX) is that they are continuously thinking about these things. Elon pushes this culture (though sometimes haphazardly). So effectively, the design and prototyping process, despite being on an urgent schedule, carries through more generations of engineering thought and redesign than would be typical in most inertia-laden organizations.
(Some of the fascinating information about the Cybertruck engineering illustrates this culture, where the product is much more than a risky cosmetic design dropped over a well-proven collection of existing company technology.)
in the AV world we see the example of Waymo who has a very capable but quite expensive platform, backed by a very capable but quite expensive engineering team using quite expensive infrastructure to develop a low-volume but impressively performing L4. (Yes I know that in detail, Waymo is not on their "first generation", but in the macro industry-deployment sense this still fits the point).
Tesla, rather infamously, has already cut corners that make people nervous or derisive, eschewing Lidar, advanced radar and even common proven hardware like basic ultrasonics. Effectively, they are trying to skip the expensive first generation and move to the lower cost second generation, if not yet a super-highly refined third-generation hardware/software form that we may see in a few years.
This is quite interesting, because it suggests that someone had the insight that it was more important to have it on every car and to get large data back from a possibly inadequate yet affordable hardware platform - that BTW as people often forget, needed to make high volume sense as designed circa 2017. It was not the typical over-engineered first generation hardware; it was a set of entirely unprecedented and audacious standard equipment on an affordable volume production car.
I try to always keep this in mind, and it's why I've focused my complaints about the cameras more on the placement (likely inherited from the prior Mobileye AP), i.e. on their viewpoint geometry, not on theories of their number being way too few nor on their inadequacies compared to exotic first-generation sensors that could never have been incorporated into the production car.
diplomat33
Average guy who loves autonomous vehicles
I get all that. And yes, it is possible Waymo is that "1st gen" that solves L4 but it is too expensive to really commercialize at scale and then we see a "2nd gen" that solves L4 later but does it in a cheaper, more scalable way. Although, I don't think it is totally impossible that Waymo is able to bring costs down on their Waymo Driver to where it does become scalable on consumer cars in the not too distant future. After all, we've already seen the Waymo Driver drop in cost 2x from the 4th Gen to the current 5th Gen and I suspect the hardware on the Geely vehicle will be even cheaper.
But did Tesla go far too the other way (reduced sensors too much) where they are not able to reach a meaningful L4 at all? That is possible too. It is great to focus on a scalable, cheap solution to autonomous driving but you still need to be able to do safe, reliable autonomous driving. I think that Tesla will need to self-correct at some point by adding a bit more sensors (cost) back in because I don't think the current hardware (HW3) is adequate for safe, reliable L4. The cameras are too low res. It needs more cameras and better placed cameras. There is no self-cleaning for adverse weather conditions. And there is no sensor redundancy. There is certainly no way I would ever trust HW3 to do driverless.
But I think the ultimate solution to autonomous driving will be something in between Tesla and Waymo. It will have more and better sensors than Tesla but not go as far as Waymo. That is one reason why I like the Mobileye approach. They are sort of the best of both (Tesla and Waymo). They start with a scalable solution to consumer cars with vision-only but add affordable radar and lidar to reach L4. For example, if you look at Mobileye Chauffeur, it adds 5 radar and 1 lidar, not as much as Waymo but more than Tesla. Mobileye also adds scalable maps and RSS that make it better than Tesla while not being as expensive as Waymo.
In conclusion, I think we might think of three development "generations" of autonomous driving:
- "Gen 1" is high safety/reliability, high cost. That's Waymo. It is highly safe/reliable L4 but very costly.
- "Gen 2" is low safety/reliability, low cost. That's Tesla. It is less safe/reliable (requires driver supervision) but it is cheaper.
- "Gen 3" is high safety/reliability, low cost. We don't know who that is yet. But this is the ultimate goal because we want very high safety and very high reliability but also at low cost.
Ultimately, I think we will get to that "gen 3" stage of autonomous driving that is high safety/reliability as well as low cost. Long term, I am very optimistic about autonomous driving. It has taken time but as tech improves and costs come down we will eventually reach "gen 3".
Similar threads
- Article
