willow_hiller
Well-Known Member
Have you watched Tesla's presentation from CVPR this year? It also touches on end-to-end:
-
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
Have you watched Tesla's presentation from CVPR this year? It also touches on end-to-end:
Yes I have. I didn't find it very interesting (not much new). Are they going to toss out their current modular architecture and start from scratch yet again? Perhaps in a few years? E2E isn't a magic bullet, and it comes with lots of problems rn.
This was the frank and good discussion I was referring to:
Full disclosure: I'm a long time secret admirer of Sergey Levine.
All talks are available here:
Last edited:
Knightshade
Well-Known Member
That's the fundamental issue with solving a problem nobody has solved yet and why all predications are so garbage on all sides.
Nobody, and I mean nobody, actually knows what is needed to solve it until someone does.
Tesla has thought they knew what at least 3 or 4 times now with various HW and SW approaches and versions? And been wrong every time- they keep eventually finding limits they didn't expect and having to go back to the drawing board.
Maybe the NEXT one will be the one. Or maybe the next 5 won't. And each takes years, and until one works you don't know which one will.
And it's not like anybody else taking different approaches has "solved" this yet either--- though they have deeper, but narrower, offerings in the space (L4 in parts of a few specific cities for example, or L3 on specific highways at low speeds)
I agree. And If only Tesla wasn't busy fighting their crap hardware they might actually have a chance to contribute to the field instead of selling dreams and pumping that stock.
I've said for a very long time (in my local Tesla UG forum) that FSD city streets is a hopeless task right now. Highway L3 is on the other hand definitely in reach.
But Tesla can't have that because they already sold the other thing. They instead ship a (pretty useless) supervised forever* city streets FSD and start pumping the stock using pipe dreams like humanoid robots, which is a problem several magnitudes harder.
*) my car is four years old now, and I don't think it matters if i keep it four more. I still won't get an autonomous car.
Last edited:
I wouldn't, given it works fine on my car. I think the post was informative. I've seen it posted a million times that the sensor costs dollars, or even cents, but didn't realize it costs considerably more, which changes the cost to benefit a lot. If it's a $200+ option, I have serious doubt people would opt for it.
Most of the issues stem anyways from not having a sensitivity adjustment, not necessarily the detection itself. Even the rain sensor cars have one, because automakers realize you can't have one size fits all, as Tesla is attempting.
Knightshade
Well-Known Member
FWIW I think L5 is WAY harder than human robots doing specific tasks.
Generalize AI in a humanoid robot? Sure, way harder than either.
But I fully expect Optimus to be doing actual work not just demos long before I expect to see Tesla L5 robotaxis.
SF police can't ticket Waymo for moving violations, that's one of complaints SFMTA has. So even if a cop saw it, no such tickets or reports would exist! They can only ticket for equipment, for example a broken light.
When a Robotaxi Gets a Ticket, Who Is Accountable?
Cruise and Waymo can get parking tickets, but not most moving violations.sfstandard.com
That's interesting. Seems like a quirk of local law which should be addressed.
Around here if a FexEx vehicle (for example) went through a red-light and the officer got the plate but was unable to give them a ticket, it would be sent to the registered owner of the vehicle (FexEx in this example). How does SF handle fleet vehicle incidents (like taxi, delivery, etc.) which run from the scene (purposfully or accidentally) so the specific driver cannot be identified by the officer?
Since Waymo vehicles have plates they would get tickets. Taking points off the license is a far more challenging item.
For sure, we've had specialized robots for 40 years. Industrial robot - Wikipedia
Some humanoids like BD (now Hyundai) Atlas, Honda, etc. Lots of research ones too, like
Dennis was brought in to help Tesla build the first version btw.
However, the only (real) reason to pick the humanoid form is for social acceptance and possibly the ability to navigate in a world designed for humans (like stairs). Tesla's motives seem to be mainly stock pumping and pulling "an FSD" all over again, as the sw problem is likely decades away for it to make any sense to build a humanoid for use in a factory.
I'm a lot more impressed by DeepMind's recent innovations in software, and Everyday Robot's work with robotics in office spaces than the useless Optimus-without-a-brain piece of metal.
RoboCat: A self-improving robotic agent
Robots are quickly becoming part of our everyday lives, but they’re often only programmed to perform specific tasks well. While harnessing recent advances in AI could lead to robots that could...
Home | Everyday Robots
Born from X, and working alongside teams at Google, we’re building a new type of robot. One that can learn by itself, to help anyone with (almost) anything.
everydayrobots.com
Last edited:
Knightshade
Well-Known Member
I think there's an argument to be made that humanoid lets you use them in places that are, right now, designed for humans without having to redesign your entire business around a non-human form factor doing the work.
I'm not saying how GOOD an argument that is- and I think it'd be a complex debate whose answer would vary by industry too- and probably is getting way far afield of the topic here--- but I recognize it as AN argument other than just social acceptance.
Yeah, I was editing my post just now regarding that. Stairs etc. But again, the software is likely decades away, and BD already has a brain-dead humanoid that is too expensive for most applications. I think something like the Everyday Robots approach can work in offices and in factories. That form factor is a good trade off imho. Plastic for office/domestic and metal for industrial.
Last edited:
In California, generally vehicle owners are not liable for moving violations. Even when owner is issued a ticket with evidence (like for example a speeding or red light camera), the prosecution has to prove you were in the vehicle, as such the image of the driver must be clear enough in such cases to identify the driver. There is currently no legal mechanism to make owners liable when there is clearly no driver in the car.
They can however ticket cars for non-moving violations, given existing parking laws can be cited and does not require the driver to be present in the vehicle. As such, all of the reports you see are of this variation (blocking intersection or road, double parking, blocking emergency vehicles, etc).
Last edited:
diplomat33
Average guy who loves autonomous vehicles
No better opportunity exists on earth than a two hour wait for chargers in a city the size of Houston.
Reminds me of when taxies took advantage of free unlimited supercharging, prompting Tesla to eventually put restrictions on commercial use (although not retroactively).
This time however it doesn't appear to be directly financially related, probably operation wise this is the only way they can keep cars running continuously at this moment (even with a home base it might not have fast charging).
Not sure if charge networks would necessarily step in either if users complain to them.
Last edited:
If there is that much demand the charge networks should actually be making money, which would make them want to add more locations. (Though I suspect a lot of them are waiting for NEVI funds to get granted, not that that would likely even apply in an area where Cruise is charging.)
diplomat33
Average guy who loves autonomous vehicles
Ghost Autonomy released their first demo of their system.
Description: Ghost was built to generalize with universal AI, no HD-Maps, and fully auto-labeled training data. In this video Ghost drives autonomously on a road it’s never seen before – Interstate 15 from California to Las Vegas, NV, traveling 140+ miles with no safety driver interventions.
Description: Ghost was built to generalize with universal AI, no HD-Maps, and fully auto-labeled training data. In this video Ghost drives autonomously on a road it’s never seen before – Interstate 15 from California to Las Vegas, NV, traveling 140+ miles with no safety driver interventions.
diplomat33
Average guy who loves autonomous vehicles
Zoox starts deployment of their no steering wheel robotaxi in Las Vegas. They are starting with a 1 mile loop at 35 mph. Definitely not impressive but they promise to scale up in the coming months.
Knightshade
Well-Known Member
Meanwhile The Boring Company still can't use basic autopilot on the Teslas in their one-lane tunnels in Vegas.
More cases of stalling after SF pride parade and Waymo cars can't figure out how to handle traffic. Reports of stalling went up 3x in recent two months (May and June) compared to prior two months (March and April) Articles also mention Thursday vote by CPUC on expanding to 24/7 service.
jalopnik.com
www.google.com
Two Waymo Cars Block San Francisco Traffic Again As Robotaxi Stalling Incidents Rise 300 Percent
The self-driving cars were left motionless by heavy San Francisco traffic due to the Pride Parade and Giants game on Sunday.
jalopnik.com
2 Waymo self-driving cars stall at SF Pride Parade street closures during heavy traffic
Two self-driving Waymo cars stalled out in the middle of two separate intersections just blocks away from each other in downtown SF following the Pride parade.
www.google.com
Doggydogworld
Active Member
CPUC hearing postponed to July 13. As for stalling, I wonder if they're driving 3x as much, stalling 3x as often per mile or if reporting stalls has just become a thing (as occasionally happens with 'unintended acceleration').More cases of stalling after SF pride parade and Waymo cars can't figure out how to handle traffic. Reports of stalling went up 3x in recent two months (May and June) compared to prior two months (March and April) Articles also mention Thursday vote by CPUC on expanding to 24/7 service.
Two Waymo Cars Block San Francisco Traffic Again As Robotaxi Stalling Incidents Rise 300 Percent
The self-driving cars were left motionless by heavy San Francisco traffic due to the Pride Parade and Giants game on Sunday.2 Waymo self-driving cars stall at SF Pride Parade street closures during heavy traffic
Two self-driving Waymo cars stalled out in the middle of two separate intersections just blocks away from each other in downtown SF following the Pride parade.
Similar threads
- Article
