EVNow
Well-Known Member
NVidea had a white paper on e2e NN for FSD a few years back !
-
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
DanCar
Active Member
Layoffs in May and hiring heavily 6 months later."Robot delivery startup Nuro announced plans to layoff a portion of its workforce and to pause its commercial operations as it pivots to more research and development. The news comes amid a broader set of financial challenges for the burgeoning autonomous vehicle sector."
May 2023
/cdn.vox-cdn.com/uploads/chorus_asset/file/23161324/Hero_Image_2__Nuro_Groceries.jpg)
Nuro plans for more layoffs as the AV sector’s economic woes deepen
The AV sector is experiencing a lot of financial headwinds.www.theverge.com
Work at Nuro
We’re building an autonomous delivery service that will make neighborhood roads safer and reduce emissions. Join our team!
www.nuro.ai
The takeaway was the pivot from commercial to r&d.Layoffs in May and hiring heavily 6 months later.
Work at Nuro
We’re building an autonomous delivery service that will make neighborhood roads safer and reduce emissions. Join our team!
Tam
Well-Known Member
Say goodbye to TuSimple.
It couldn't find a buyer to take it over so it shuts down in the US.
techcrunch.com
It couldn't find a buyer to take it over so it shuts down in the US.
Another blow for self-driving trucks as former industry leader abandons the US | TechCrunch
When TuSimple went public in 2021 it was flying high as the leading self-driving trucks developer in the United States. Now — after a string of internal
techcrunch.com
EVNow
Well-Known Member
Is it really worthwhile to pursue robotrucks .... what is the labor cost as a percentage of overall cost ? Unlike urban robotaxis (majority of cost is labor) - the operational aspects are difficult too. Imagine trucks getting stuck in nowhere land ... like Cruise.Say goodbye to TuSimple.
It couldn't find a buyer to take it over so it shuts down in the US.
Another blow for self-driving trucks as former industry leader abandons the US | TechCrunch
When TuSimple went public in 2021 it was flying high as the leading self-driving trucks developer in the United States. Now — after a string of internal
Tam
Well-Known Member
I think it's an excellent application.
Machines love doing repetitive tasks: the less thinking, the better.
Truck routes from a company depot to a depot are not indefinite. The routes are pretty much unchanging. Thus, it's all set once it learns how to drive in one particular route! No new learning is needed.
GM Cruise got stuck because it had to learn strange places, new routines, routes, and behaviors:
The Cruise on the left pondered whether it should yield to the one on the right. The one on the right was thinking the same. Too much thinking, and the whole neighborhood got stuck with a bunch of Cruise cars!
EVNow
Well-Known Member
Actually machines don't feel anything
Seriously - even though truck routes are simpler (and repetitive) - it doesn't mean you can make a particular truck learn that one route or have different software for different routes. You need a single software that handles all the routes.
Besides, Its not just the route - its the ever changing traffic and weather. Think of all the variations FSD has to handle just in your one commute route.
Finally - its a business question. Is it economically worth it. Will they ever recoup the billions spent developing.
ps : Besides the whole idea of trucking is stupid - should be using trains.
ps. the ride-hailing thing is as stupid. You should be building public transport.
I like trains but ..no. They don't work for modern freight needs. They are fixed and not dynamic which is the USA differentiator. For long long distance runs from one coast to another, super, for commodities that can sit for weeks, super. Just try to do anything freight wise with trains, it is horrid. You have to book weeks/months in advance. You have to send a truck to a client, then the truck to some rail yard, deal with brokers, do it all backwards at the other end. Horrid. You can't build a logistics hub beside a rail line because rail lines aren't where the logistics hubs need to be. Clients are almost never by rail so there are 2 trucks and rail at a min. Usually 2 rail companies (so therefore requiring a broker) and 2 trucks (or more).Actually machines don't feel anything
Seriously - even though truck routes are simpler (and repetitive) - it doesn't mean you can make a particular truck learn that one route or have different software for different routes. You need a single software that handles all the routes.
Besides, Its not just the route - its the ever changing traffic and weather. Think of all the variations FSD has to handle just in your one commute route.
Finally - its a business question. Is it economically worth it. Will they ever recoup the billions spent developing.
ps : Besides the whole idea of trucking is stupid - should be using trains.
With a truck I can call a broker, picked up at a client site, in hours or a few days it is delivered to the client. Done..that's it. One bill. For a larger company it is even easier, arrange trucking directly. Or larger yet own the truck and ship where and when you need it. 80% is by truck and when new cities emerge and freight destinations change trucking can adapt. To send anything from a medium sized city to another medium sized city on either coast (basically across time zones) is going to mean that the train car is going to go to a major hub like Chicago or Kansas City and get routed from 1 company to another and get delayed, lost, broken into, etc. By the time it's all said and done the 2 day truck ride is a 2 week train ride with a complex set of interactions (and potential failure) to manage and monitor.
You can hire companies like JB Hunt to ship a container and they'll figure out if they can send it part way via train as they have freight forwarding agreements with rail yards. Maybe instead of rerouting something from LA to Pittsburgh via Chicago by changing trains they just pull the container in Des Moines and ship by truck to Pittsburgh (for example). They'll figure that out and price it and keep most of the extra profit but pass a few dollars back. It's not a huge savings.
This is why freight has been shifting from train to truck for decades. Trains are also at capacity in many areas so we have chokepoints. The economics aren't there to expand and NIMBYs don't want a new massive train depot in the backyard. Trains are great for bulk commodities (coal, grain, etc) and regular scheduled trips from a port to a distribution center but no so good for anything else.
EVNow
Well-Known Member
Yes - I understand the current system is setup for trucks. In the long run - its better to optimize the system for trains. Electrified trains is the most climate friendly way to move goods.
diplomat33
Average guy who loves autonomous vehicles
Unedited video of Wayve autonomous driving in London using their embodied AI end-to-end approach:
At 5:27, I think the car should have slowed more and yielded to the 2 older women crossing the street. I feel like it zipped by way too close to them. If they had not stopped, that could have been a serious collision.
At 5:27, I think the car should have slowed more and yielded to the 2 older women crossing the street. I feel like it zipped by way too close to them. If they had not stopped, that could have been a serious collision.
Last edited:
EVNow
Well-Known Member
Sorry - can't let that slip through.
It is only unedited if you ignore that they chose exactly which route and exactly which video to show. All other routes and takes were edited / censored.
If its not 3rd party, with no strings attached - its propaganda / marketing /PR.
diplomat33
Average guy who loves autonomous vehicles
When I say unedited, I mean there are no edits or cuts during the video, ie the video is a continuous unedited drive from A to B. In contract to Tesla's "paint it black" video which took several pieces and edited them together to look like a continuous drive when it was not.
And I never said that the video was not marketing/PR. But the video is a sample of end-to-end autonomous driving with some interesting scenarios. Considering, how much this forum is discussing end-to-end approach, I thought a video showing some end-to-end autonomous driving would be worth posting.
EVNow
Well-Known Member
Sure - that’s why rated your post as informative.
Unedited and edited videos only differ in how many times the PR is willing to try before getting a take that they can publish
Doggydogworld
Active Member
I didn't think Paint it Black was spliced, but that they attempted something like 30 runs before they got one without intervention.
diplomat33
Average guy who loves autonomous vehicles
Doggydogworld
Active Member
Thanks, I'd never seen that. I knew about the multiple attempts, but thought the "spliced" claims were just TSLAQ FUD. I guess the lesson learned is to always doubt Tesla and never doubt TSLAQ (j/k @EVNow , put down your flamethrower
EVNow
Well-Known Member
Demos are just that ... you want to show something (that might work in the future). Just see the controversy now surrounding Google AI demo.Thanks, I'd never seen that. I knew about the multiple attempts, but thought the "spliced" claims were just TSLAQ FUD. I guess the lesson learned is to always doubt Tesla and never doubt TSLAQ (j/k @EVNow , put down your flamethrower
In the old days - people used to do live demos on stage. The chances of something going wrong was high ... so started recording the demo and showing on stage. Enter fake demos.
If someone wants to prove their FSD works - give the car to an independent journalist for a week - don't tell him which routes to follow and let him show how it worked. Ofcourse, the best way to figure out is what Tesla is doing now - send it out to a million users to test.
Google Admits Gemini AI Demo Was at Least Partially Faked
Google has a lot to prove in its efforts to catch up with its fierce AI competition. But given the latest news, the company seemingly can't stop tripping over its own feet. Earlier this week, the tech giant announced Gemini, its most capable AI model to date, to much fanfare. In one of a series […]
ps : Do I think Tesla would fake demos - absolutely !
Similar threads
- Article
