-
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
willow_hiller
Well-Known Member
The context was "Waymo is secretive." Not in comparison, but in the abstract. And they've sued the California DMV to keep their crash data as secret as possible. Waymo wins bid to keep some of its robotaxi safety details secret
Research papers are not necessarily indicative of openness. As all businesses will, they'll present things publicly when it benefits them to do so.
Dude, Tesla doesn’t report ANY meaningful KPI:s around FSD.
fasteddie7
Active Member
I've seen so many posts and videos on how bad park assist was, that when I just got it on my HW4 S I was expecting a dumpster fire, in some ways it turned out better (no curb transition to grass from a parking lot was visualized perfectly) the only issue I had was if I place an object close to the car when it's off, it doesn't know it's there when it starts.
diplomat33
Average guy who loves autonomous vehicles
Cruise finally adds Android support to their ride-hailing app.
powertoold
Active Member
How can anyone watch a video like this and not understand what's happening, considering Tesla's pace of improvement. I keep saying Waymo's driving performance will seem primitive by this time next year:
As it relates to how the tech works however, how it is rolled out, and things like regular release notes, I think people have a much better grasp on Tesla's system than they do on how Waymo's tech works and is rolled out. Maybe people were busy, but for example I asked about evidence on how Waymo's tech currently works in regards to ML in situations where hand coding seems to make way more sense, and although academic papers were alluded to in response, links were not posted. Instead there was back and forth based on more general public statements made by Waymo, but nothing in terms of technical details.
Instead for Tesla, due to very public tracking of releases, release notes, drivers who can consistently compare differences in releases (and post them on YouTube for everyone to see), hackers like green (which Tesla seems to fund/allow via their bounty program) that look under the hood, we get a relatively good understanding on how Tesla's system works. Even for people like me, that only occasionally follow up on it, I can see how the system evolved and in what parts they have introduced NN vs hand coding for example. It's much less of a black box.
Here is a video showing Waymo, Cruise, and Tesla FSD all starting in the same place at the same time going to the same destination:
All three took a different route.
Technically Tesla FSD "failed" because it required one intervention, tapping the accelerator, but other than that it was the first to arrive.
All three took a different route.
Technically Tesla FSD "failed" because it required one intervention, tapping the accelerator, but other than that it was the first to arrive.
Seems like finally this is set up after people calling for it for so long. Haven't watched it through yet, so don't know the result yet, but I'm sure there will be accusations of things being rigged to favor the Tesla, especially given the event was set up by someone some people here call a "shill", but it'll be an interesting watch regardless.
Edit: just looked at the initial bit, it looks like the Tesla chose to get on the freeway, exactly the route that I would expect it to choose just based on my own experience with Tesla navigation (I don't have FSD or even EAP, just AP). So its route choice is more like how real human drivers would drive in their own cars.
Edit2: I personally don't prefer freeways for local routes (wastes energy), but the next choices of travelling Franklin and Pine (both efficient thoroughfares that traffic lights favor the traffic travelling through) is the same choice I would make.
Edit3: he did one accelerator press to get past a yellow (which he definitely should not have done). I would probably take about a minute off his time for that, but definitely not fair to give a time advantage by intervening.
Last edited:
Is he not the definition of that?
His videos are curated to only show the best of the software, but then he advertises it like it's common. He also sets it up to lead others to believe it's completely hands-free.
The Webster's definition is an accomplice of a hawker, gambler, or swindler who acts as an enthusiastic customer to entice or encourage others.
That's exactly what he's doing.
Without watching I guess Tesla "won" because it could take the interstate knowing Waymo and Cruise cannot? Is that close?
powertoold
Active Member
It doesn't matter that Tesla "won" this race. What matters is the underlying technology and what it has taken to get here and where it'll be in the future.
People who follow my train of thought would think: Tesla won by leveraging massive computation. Leveraging computation to solve problems is the way of the future, not hand-coding and manually driving through the city, mapping out everything mostly by human hands. Like George Hotz, I don't understand why people can't see that Tesla FSD is way ahead of everyone else in the US. Is this an IQ test or are people just hating on Elon / Tesla just for being "late"?
People who follow my train of thought would think: Tesla won by leveraging massive computation. Leveraging computation to solve problems is the way of the future, not hand-coding and manually driving through the city, mapping out everything mostly by human hands. Like George Hotz, I don't understand why people can't see that Tesla FSD is way ahead of everyone else in the US. Is this an IQ test or are people just hating on Elon / Tesla just for being "late"?
diplomat33
Average guy who loves autonomous vehicles
Mobileye CTO Prof. Shai Shalev-Shwartz was at EcoMotion 2023 and had some interesting things to say about autonomous driving progress.
A few quotes:
“By now there is evidence that self-driving is possible,” said Prof. Shalev-Shwartz. “We see self-driving cars in the streets of San Francisco and Austin. So I think that we are beyond this obstacle. The problems are different.” Those problems are the same that Shalev-Shwartz and his colleagues laid out in a research paper six years ago, namely: safety and scalability.
As Shalev-Shwartz put it: “There is no way to guarantee absolute safety. So the question is, if we cannot guarantee absolute safety, what can we guarantee? And how can we have a language that enables us to talk about exactly what is ‘safe’ and what is ‘not safe’?” That’s what RSS seeks to define, he noted. “If everybody would act according to RSS, then there would be no accidents at all.”
“We view AVs as a marathon and not a short sprint. We are trying to be very honest about what the technology is capable of, and what the technology is not capable of.” "We believe that the right way to win a marathon is to start with the foundations and then make incremental changes, incremental improvements, while having a solid business all the time, rather than trying to make promises or to go all the way to something that you cannot hold your breath for long enough,” said Shalev-Shwartz
Ultimately, Prof. Shalev-Shwartz concluded, “I think the future is going to be very, very interesting. And I think AVs are part of the future and are here to stay, and we will all enjoy them in the future.”
Source: Mobileye CTO: “AVs Are Part of the Future and Are Here to Stay” | Mobileye Blog
Full Session:
A few quotes:
“By now there is evidence that self-driving is possible,” said Prof. Shalev-Shwartz. “We see self-driving cars in the streets of San Francisco and Austin. So I think that we are beyond this obstacle. The problems are different.” Those problems are the same that Shalev-Shwartz and his colleagues laid out in a research paper six years ago, namely: safety and scalability.
As Shalev-Shwartz put it: “There is no way to guarantee absolute safety. So the question is, if we cannot guarantee absolute safety, what can we guarantee? And how can we have a language that enables us to talk about exactly what is ‘safe’ and what is ‘not safe’?” That’s what RSS seeks to define, he noted. “If everybody would act according to RSS, then there would be no accidents at all.”
“We view AVs as a marathon and not a short sprint. We are trying to be very honest about what the technology is capable of, and what the technology is not capable of.” "We believe that the right way to win a marathon is to start with the foundations and then make incremental changes, incremental improvements, while having a solid business all the time, rather than trying to make promises or to go all the way to something that you cannot hold your breath for long enough,” said Shalev-Shwartz
Ultimately, Prof. Shalev-Shwartz concluded, “I think the future is going to be very, very interesting. And I think AVs are part of the future and are here to stay, and we will all enjoy them in the future.”
Source: Mobileye CTO: “AVs Are Part of the Future and Are Here to Stay” | Mobileye Blog
Full Session:
Well, company promo videos (which are heavily edited and even more curated than his) is frequently posted here as evidence of prowess, but we don't generally use as negative a term as "shill" to describe it.
I appreciate that at least he is willing to show uncut footage (even if it might be his best runs) and show the wheel and pedals to indicate his interactions. My understanding is he only did this because people were accusing him of secretly intervening in his videos (examples I saw upthread). And from what I see in the video it is "hands-free," which I think Tesla's ultimate goal is to get to with the cabin camera (others have already done that with L2). This is different from "eyes-off" obviously.
That's amazing... if Waymo was actually 24/7 driverless across the entire LA metro area.
What is their LA coverage map? Who's getting drives right now?
Do you realize the LA metro area is 33,954 square miles? If Waymo this year covers say 100 square miles, they will have to scale that up 338 times to cover the full LA metro area?
So yeah I think your concept of diminishing time to ramp up a new area is legit. I just think the real areas being covered are actually quite small.
It's refreshing to hear realistic talk about current capabilities and shortcomings as well as hope for the future versus nonstop koolaid talk.
Tesla seems to have the static problem of self-driving locked down at this point in a cheap and economical way. I always watch the visualizations and every roadway edge and curb both visible and outside line-of-sight or occlusions is super dialed in and stable now, both onroad and in adjacent parking lots and driveways, and turns have gotten super smooth and confident. Even in poor weather/heavy rain. I think they're doing a combination of vision and crowdsourced data from the fleet (according to @greentheonly). But no hand-turned HD map or super precise LIDAR localization going on obviously, just course lane counts and so forth.
Control and planning, and dynamic problems are much less mature. But it seems like you really don't need HD maps and LIDAR localization for the static problem of driving, that seems clear.
This is what I mean.
The tesla warped logic.
Simply amazing.
I don't know about yall, but continuing conversation with these people are complete waste of time.
You're wasting your time with these people. They don't care about your facts. Logic and reason is not involved.
Similar threads
- Article
