I just uploaded a clip of a video I took a couple of months ago from the back seat of an otherwise empty Cruise Bolt EV driving me around at night. In the span of less than a couple of minutes it gracefully and quickly handles 3 double-parked cars on different street blocks. It seems pretty clear to me that this wasn’t remotely controlled by a human.
-
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
Terminator857
Active Member
My estimate was for all of the U.S. When do you think you'll be able to do driverless? Will it be Waymo or Tesla?
Terminator857
Active Member
You have to multiply that by the number of times the roads are used.There are about 3 million miles of paved road in the U.S. and .000001% of that is 160 feet.
Daniel in SD
(supervised)
That's only 30 miles. It's higher than that too!
I think you all are confusing % and decimal fractions.
Terminator857
Active Member
3.2 trillion miles per year. So if we guess that waymo is driving 32,000 fully driverless miles per year, that comes out to: 1E-8.
Quite a ways to go.
32,000 / 3.2 trillion = 1E-9.
If convert that to percentage that would be .0000001%
Is the math good?
Last edited:
EVNow
Well-Known Member
diplomat33
Average guy who loves autonomous vehicles
Me too. I am saying Tesla has deployed driverless vehicles in 0% of the entire US. Waymo has deployed driverless vehicles in about 0.001% of the entire US. You need to measure actual deployment of driverless vehicles because deployment is what really matters, not how many cars are testing FSD.
Well, solving driverless for the entire US with vision-only is monumentally hard even with a lot of data. Furthermore, I live in a small town in Indiana. There are not as many Teslas in my area than in say California. The data is probably a lot sparser in my area. So it might take longer for Tesla to "solve driverless" in my area than in other areas that have more data. There is also the possibility that the current hardware is not good enough for driverless. On the other hand, my area is not prime market for Waymo. Waymo might deploy driverless in most US cities and ignore my area. So it is kind of hard to say who will be first in my area. There are many variables. It's basically a race between can Tesla "solve FSD" everywhere with just vision-only fast enough versus can Waymo deploy their proven driverless to more areas fast enough? Maybe Tesla will solve driverless in 2 years so I get driverless before Waymo. Maybe Tesla cannot solve driverless and Waymo does scale and Waymo gives me driverless first. Maybe Tesla takes 10 years to solve driverless in my area but Waymo ignores my area so I still get driverless first on Tesla. But I think Waymo has a much better chance of "solving driverless" everywhere than Tesla. I say that because Waymo has shown they can do driverless, they just need to expand it. Tesla has not shown that they can do driverless yet.
I will say I think Tesla fans get wrong about Waymo, that waymo has to "specialize" FSD to each new area so Waymo will take way too long to scale to the entire US. In fact, Waymo's FSD is generalized. Waymo has also confirmed that the learning is shared across the fleet. So Waymo's scaling will be exponential. As they learn, they will expand faster and faster. So it won't take decades to expand to most US cities. My guess is that Waymo will do driverless in most major US cities in 5 years. In fact, I think Waymo, Mobileye, Cruise and Argo will have widescale driverless in 5 years. So, if Tesla does not "solve driverless" in 5 years, they will lose the FSD race.
Last edited:
voyager
Active Member
Investors have sunk more than $100 billion into building cars that can drive by themselves. Yet despite a decade and a half of development and untold miles of road testing “we are seeing extraordinary amounts of spending to get very limited results”.
"Newcomers like Wayve and Autobrains have ditched HD maps entirely. Wayve’s cars have GPS, but they otherwise learn to read the road using sensor data alone. It may be harder, but it means they are not tied to a particular location. For Kendall, this is the key to making driverless cars widespread. “We are going to be slower to get into our first city,” he says. “But once we get to one city, we can just scale everywhere.”
The big new idea for making self-driving cars that can go anywhere
The mainstream approach to driverless cars is slow and difficult. These startups think going all-in on AI will get there faster.
www.technologyreview.com
Daniel in SD
(supervised)
I'd like to see data on the cost of HD mapping. What is it? Has it fallen? Is it likely to fall in the future?Investors have sunk more than $100 billion into building cars that can drive by themselves. Yet despite a decade and a half of development and untold miles of road testing “we are seeing extraordinary amounts of spending to get very limited results”.
"Newcomers like Wayve and Autobrains have ditched HD maps entirely. Wayve’s cars have GPS, but they otherwise learn to read the road using sensor data alone. It may be harder, but it means they are not tied to a particular location. For Kendall, this is the key to making driverless cars widespread. “We are going to be slower to get into our first city,” he says. “But once we get to one city, we can just scale everywhere.”
The big new idea for making self-driving cars that can go anywhere
The mainstream approach to driverless cars is slow and difficult. These startups think going all-in on AI will get there faster.
Everyone seems to think it's a problem but I've never seen it quantified.
Cost of HD mapping should end up 0: All cars need to be able to function without a HD map, but HD maps can increase their confidence. When cars enter non-mapped areas, they can start building a map while driving and send it back to the home base. After enough cars have shared their maps, these can be combined with confidence and shared.
This is better than what humans can do. When you drive to an unknown area for the first time, you learn. Driving there 100th time, you have built a lot of confidence and drive better. What we are not able to do is to share those learnings (memories) with others in scale.
This is better than what humans can do. When you drive to an unknown area for the first time, you learn. Driving there 100th time, you have built a lot of confidence and drive better. What we are not able to do is to share those learnings (memories) with others in scale.
diplomat33
Average guy who loves autonomous vehicles
If they think this, they are sorely mistaken. It is not going to be that easy. Every city has new challenges and new edge cases. One city is not enough to scale everywhere. What is going to happen is they are going to "solve one city" and then realize it does not work well enough in another city. So they will need to do more NN training with more data to get it to work in the other city, then they will realize it does not work well enough in city #3 and #4 and needs more training etc... It will be much harder than they realize. The fact is that there are no easy shortcuts to "solving FSD".
diplomat33
Average guy who loves autonomous vehicles
This is good news. This new regulation will allow companies like Mercedes to relax the limits and allow full highway speed L3. I wonder how long before we see the Mercedes L3 system go up to 130 km/h?
They also now allow automated lane changes as well. Mercedes did say their current system can support higher speeds, but the rule currently doesn't allow. I do wonder if they will update their current system to keep pace with the rule or would they only offer it in newer model vehicles like legacy automakers like to do. Lol
diplomat33
Average guy who loves autonomous vehicles
Yeah like GM did with Super Cruise. If Mercedes is smart, they should do OTA updates IMO.
Automated lane changes should make motorway use of NoA in the UK (and other parts of Europe) much nicer. The indicator going on will provide an audible warning and allow the driver to check all is OK. More restful than having to keep checking the display for overtaking notifications and then confirming.
EVNow
Well-Known Member
If Mercedes was smarter they would have kept their Tesla shares and made a killing. They would have made more money than they make on cars. Same for Toyota
From their view, relying on HD maps is the shortcut. If the end goal is a car that can drive on any public road a human can, then eventually it'll need to pretty much be able to handle all situations you throw at it, even in areas without an HD map. The idea is there will be less edge cases that the system won't be able to handle vs having to do maps (which has a ton of repetition, even though essentially overall the road structure is different for every road). Of course that idea might be wrong (could be possible HD mapping the entire world's roads is an easier task than developing an AI that can handle all roads thrown at it), but it's unknown at the moment.
Obviously if the end goal is simply fleet robotaxis, that's a bit different. For that, you will have to expand city by city anyways due to a centrally owned fleet, so even spending time to build maps for every city isn't really a big difference.
It that same decade and a half how much was spent on the funerals for those who died in car crashes? How much in lost wages? How much for the injuries to the millions that didn't die but were injured?
Similar threads
- Article
