-
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
No speed listed for the crash, only speed limits of the roads (highway and the side road).
More recent image from Bing's "Bird's Eye" view shows the construction and lack of lights.
Yeah, from the image above it looks like they've removed those street lights while construction is ongoing.
Last edited:
The funny thing is that GM has been selling the made in China Buick Envision in USA since around 2016
diplomat33
Average guy who loves autonomous vehicles
It feels like I've seen that press release 2-3 years ago already...
diplomat33
Average guy who loves autonomous vehicles
Not with VW, I don't think. But there have been other companies that have signed on with Mobileye to deploy a L4 vehicle.
It's this (6 months ago):
Volkswagen ID. Buzz showcases Mobileye Drive™ at IAA 2023 | Mobileye Blog
Electric passenger van on display in Munich incorporates Mobileye's turnkey self-driving solution for autonomous mobility-as-a-service.
And then there is a slew of other "announcements that never came to anything" here Driverless Mobility-as-a-Service from Mobileye
/cdn.vox-cdn.com/uploads/chorus_asset/file/13353143/VW_Ridesharing_1.jpg)
Volkswagen, Intel, and Mobileye will launch a self-driving taxi service in Israel in 2019
An aggressive move that could push VW to the front of the pack.
www.theverge.com
It's just insane pumping afaik. Noise vs signal is massive from this company.
diplomat33
Average guy who loves autonomous vehicles
It's this (6 months ago):
Volkswagen ID. Buzz showcases Mobileye Drive™ at IAA 2023 | Mobileye Blog
Electric passenger van on display in Munich incorporates Mobileye's turnkey self-driving solution for autonomous mobility-as-a-service.www.mobileye.com
To be fair, I think that is part of the same thing. That was a showcase of the vehicle at IAA Mobility. This was a press release about the same thing.
And then there is a slew of other "announcements that never came to anything" here Driverless Mobility-as-a-Service from Mobileye
Volkswagen, Intel, and Mobileye will launch a self-driving taxi service in Israel in 2019
An aggressive move that could push VW to the front of the pack.
It's just insane pumping afaik. Noise vs signal is massive from this company.
Yes. There were big plans back in 2019 about launching a robotaxi service in Israel which Mobileye later cancelled when they decided not to do their own robotaxi service. citing the high cost of managing logistics was not worth it and they wanted to focus on developing tech.
I don't disagree that there is a big noise vs signal disparity with Mobileye. I like Mobileye but I would like to see a lot more concrete results rather than marketing videos and press releases about future plans.
diplomat33
Average guy who loves autonomous vehicles
NYC to allow AV testing with safety drivers only:
/cdn.vox-cdn.com/uploads/chorus_asset/file/3924522/city-scape.0.jpg)
www.theverge.com
New York City welcomes robotaxis — but only with safety drivers
NYC won’t permit fully driverless vehicles.
www.theverge.com
diplomat33
Average guy who loves autonomous vehicles
Motional robotaxi got a perfect score on the Las Vegas DMV driver's license test.
I know it's PR. Passing a driver's license does not mean that Motional has solved FSD. But it is still kind of cool to see the car take a driver's license test.
I know it's PR. Passing a driver's license does not mean that Motional has solved FSD. But it is still kind of cool to see the car take a driver's license test.
Last edited:
And it probably means these cars can drive better than many humans in tested areas.
Last edited:
OpenAI investing in Ghost Autonomy to bring LLMs to autonomous driving. I believe LLMs will play a big role in autonomous driving.
I called it
diplomat33
Average guy who loves autonomous vehicles
Key quote: "The path to long-term profitability was uncertain given the current funding climate and long-term investment required for autonomy development and commercialization."
Just goes to show how hard it is to reach profitable autonomous driving. Most companies can't do it because it takes huge amounts of capital, business plan as well as a lot of engineering expertise, huge training compute and perseverance. Companies may have a good idea like using LLMs or end-to-end, they may even raise some capital. But they can't go the distance.
Doggydogworld
Active Member
Tesla did it with a few cameras and a rigged video. Well, at least the profitable part.....
Daniel in SD
(supervised)
I would not short TSLA based on robotaxi doubts. I think the market already reflects the very low chance of Tesla deploying robotaxis.Hi, all --
I'm going to be posting this on the fsdtracker thread as well.
As a reminder, I'm interested in FSD 12 because I'm short Tesla because I think the stock overvalued. I'm more than willing to discuss to my position, but it isn't relevant to this current conversation, I'm only bringing this up because I don't want to be accused of false-flagging down the line. I think it is possible for people to disagree about stock valuations while remaining civil, and in fact even friendly. This isn't a "change my mind!" thing, it's more a, "Let me try to understand how other people are thinking about this" thing. I'm not looking for a hammer-and-tongs debate, just a conversation.
I'm going to try to be quantitative, but (of necessity) very sloppily so, I'm going to be using big round numbers, powers of 2 and 10.
I'm hoping to explore likelihood/timelines for an FSD robotaxi/personal chauffeur scenario. To keep the discussion manageable, let's limit the discussion to "FSD robotaxi in San Francisco." as the endpoint.
I'm basing this discussion on the fsdtracker,
where I take "Miles To City Critical Disengagement" to be the key metric.
That was a very long preamble! Sorry, let's dig in.
1: Is there any reason to not use fsdtracker #s for this conversation? My sense is that all systemic biases (not a criticism, Elias!) point in the direction of the tracker overestimating FSD's reliability, but people's experiences are so wildly variable that that luck of the draw could easily be dragging them down. Based on my social media monitoring, (which include a ton of TMC!) I find them plausible. Does anyone disagree? This isn't, "They don't match my experience, it always works for me," or,, "They don't match my experience, it never worked for me"; this is, "Do you find the *aggregate* tracker #s implausible"?
2: My read on the tracker #s is: 10.x and 111.x were about 80-90 mile/CCDE. (That's "City Critical Disengagement."). I've never actually done the calculation, that's an eyeball #. I just checked, and 12.x is at 180 miles, so about twice as good, but, as I think most participants in this TMC subsection would agree, nowhere close to "ride hail in San Francisco" levels.
3: What would be good enough? California requires you to share data and jump through hoops to run robotaxis. I'm going with 10Kmi/CCDE as the level at which it makes sense to start sharing and hoop-jumping, and 100kmi/CCDE as the point at which I'd start wanting to roll out a robotaxi service. Note these are very round numbers, they're powers of 10! So I get 2 orders of magnitude to start getting serious, and 3 to start rolling out.
4: How long might that take? I have no idea, and that's where I'm really hoping for some thoughts from the group. If reliability doubles every year, you're looking at 6-7 years before it's time to start getting serious about validation; if it quadruples, you can cut that down to 4. Assuming two years for hoop-jumping and validation, you get 6-9 years before you've got SF robotaxis. I know some people have much more aggressive timelines, but don't understand why. "Data!" doesn't cut it; 12.x presumably already embodies a good chunk of the data Tesla has accumulated to date, and that corpus is not going to double annually, so even 2x/year assumes some source of "non-data" improvement.
The above analysis is obviously from a really million miles away, very abstract POV. I'm fine with that, and would actually prefer to keep it that way, because we're very much in powers of 2 and 10 territory.
A final note about growth rates -- I've had my ups and downs as an investor, but at one point did 8x in 6 years, which consisted of a double in 4 years (not bad!) and a quadruple in 2. When some number quadruples in 2 years, you can really feel it. I actually think I could have done better than 4x, but the growth threw my sizing off. As someone (Alan?) pointed out, the trajectory of 12.x/E2E is likely to be settled very soon.
Sorry for the super long post. To sum up: Any reason to not use tracker numbers as the starting point? Any reason to think MBTF will see a greater than 4x annual increase?
Yours,
RP
Getting above 10k/mi per disengagement is not really possible. See this blog post:The Disengagement Myth.
So once (if?) Tesla ever achieves that there is no way to know the true performance of the system without running counterfactual simulations (what would have happened if the safety driver did not intervene). And only the developer can do that.
Shorting all public robotaxi/robotruck companies would have been a very good bet though! Are there any left?
Hi,, Daniel, --
Thanks for moving the conversation to more neutral grounds. Honestly, bringing my position up is a huge distraction but something I felt necessary. Maybe I'll try reframing it. I'd love to talk to knowledgeable people about the ifs/whens of Tesla achieving full autonomy. I'm using "robotaxis in San Francisco" for clarity, just to have a hard endpoint instead of talking about whether FSD makes you more relaxed or not or something like that.
I'll I'm hoping to do is to but some numbers around a conversation that is mostly filled with people shouting at or with each other.
Yours,
RP
RP,
I don't think you'll find any computer vision experts here, but here's a person that I trust that discusses the vision-only approach:
"Maybe it's theoretically possible just do it with cameras and maybe it will play out, but do you want to risk it and why?"
I personally am convinced the hardware costs will continue to drop dramatically over the coming years, so I no longer see the point in doing vision-only if you really believe you could drive 10x safer with more sensing and get to a robotaxi deployment 5-10 years earlier.
Doggydogworld
Active Member
IMHO Tesla will never do Robotaxis. Outside the back of an envelope the business model sucks. Musk has moved on. Fembot is his new love.
FSD sales to end users is much more lucrative than Robotaxi.
Empty seat is a huuuuge leap. It has to be 1000x safer than a sober driver because liability awards will be 1000x greater. TV and radio lawyer ads here blare "Hit by a company car or truck? Call 1-800-EZBUCKS" 24x7. FSD 12 without a safety driver is about 1000x less safe than a sober driver. So FSD needs to improve by ~1,000,000x.
How many years will 1,000,000x take? Can't say. The early gains come fast and easy -- fun times! Every time Tesla rewrites they hit the exciting steep part of a new curve. YouTubers gush! But improvement is not "exponential", or even linear. It's asymptotic. The fun soon ends and every inch becomes a grind.
The real question for me is if they'll allow eyes-off in limited situations, e.g. highways or even just highway traffic jams like Mercedes. That'd be a huge boon for Tesla owners, freeing up millions of man-hours. They'd still have to take liability, though, and it's contrary to the Level 5 / "One day the fleet wakes up" fantasy. I always say ignore what Musk says and figure out what makes sense. But in this case I don't know what they'll do.
Similar threads
- Article
