-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
Tesla Autonomy Day event livestream (11am Pacific 4/22)
- Thread starter Ryno23
- Start date
- Status
Man if they're able to pull it off ill be amazed, when he said robo-taxi by next year I just laughed and said no fucking way.
OP
OP
I believe him and think they will accomplish exactly what he says they will(they always do just usually late), just usually have to add 6 to 12 months to his timelines at best.
The fact they can pull data from the fleet on specific scenarios, train the network, the shadow test in real life scenarios, and revise is really impressive. I can't think of any way you could ever pull of FSD without this information. The value of this data is just priceless.
I highly recommend watching the neural network segment - skip the boring chip part
I highly recommend watching the neural network segment - skip the boring chip part
OP
OP
Yeah they probably should have closed the several hour software engineering talk with showing that lol
So how well will these systems perform when the road lines are faded or completely gone?
I get these cars are mostly designed for warm weather climates, but here the lines more or less disappear during the winter either through the paint fading or snow covering. And by spring everything has to be repainted. And many non major roads don't even get repainted for years.
I get these cars are mostly designed for warm weather climates, but here the lines more or less disappear during the winter either through the paint fading or snow covering. And by spring everything has to be repainted. And many non major roads don't even get repainted for years.
Yeah, it's pretty impressive.
It was even more impressive when they did it two and a half years ago
And that one even has fun music to go with it
If you can visually navigate it - the machine learning will be trained to navigate it using visual images. This was specifically discussed during the presentation.
Thats not full self driving, that's Autopilot with Navigate. You're still required to have your hand on the wheel and be ready to take over at any time
OP
OP
That was obviously a hard coded route not using the neural network(it didn't exist) to show what they were trying to achieve eventually. We have impressions from people that went on demo rides today so it's actually a real thing, not a shortcut, and a bit more impressive.
Did the people on the demo rides have any say in the route the car took? Or was it a fixed route that Tesla was well prepared for?
If this is going to be feature complete in less than 7 months I think we should be seeing footage of it handling weird cases and bad weather, rather than cruising through the california suburbs on a nice day.
These impression tweets are hyperbolic as hell. Light years ahead of who? They are not light years ahead of Waymo or Cruise or any of the other major players. Good demos on perfect roads in minimal traffic ain't shit. My opinion of Tesla is they are just more willing to take immediate risks in this area due to Elon's pressure on his staff to push the limits (reminds me of prime Steve Jobs at Apple) and they need to release this first to maintain profitable growth and that's why they will be first to market. Still super skeptical of a completely lidar free approach. Let's see one of these events on shitty roads in shitty weather aka wet pea soup fog. Tesla May be first to mass market eventually, but to me that just means first to have a major failure/incident and bear the burden of public and regulatory scrutiny. Guess someone has to take it up the ass from litagants first though so might as well be Tesla.
Ok seriously - did anyone actually watch the presentation? They talk about all these issues. They train the ai constantly on these outlier cases and evaluate every time someone takes over for autopilot. They then roll out to the fleet a shadow program, to test the fix and gather real time data to evaluate the fix and iterate.
OP
OP
So I think in regards to the lidar vs tesla debate, it's worth noting that if Tesla wanted to add lidar to their cars, they could have. If all the companies using lidar wanted to use Tesla solution because they realized it was better, they couldn't because they don't have a fleet of 500k customers training their network. They have no choice but to be using lidar, Tesla is the only company on the planet right now capable of doing this the way they are.... Hence Elon saying "Lidar is doomed", "Game. Set. Match" etc.
I watched the whole thing. They said that's how it worked, but the only footage they showed was of the "cut in" handling which isn't really that impressive.
The system design sounds logical and may be capable of getting to the point where it can handle anything you can throw at it within reason, but I don't think it's unfair to ask how far they are and if their target date is actually realistic.
Also, how many man hours are necessary to handle each one of those cases? How many samples of a weird scenario do they need in order to effectively tweak the NN? How much manual classification by humans is necessary? How do you catch false positives and false negatives when the consequences (in that particular instance) aren't enough to cause the driver to deactivate? How does it learn city driving from The Fleet if The Fleet doesn't do autopilot in cities?
Edit: One thing I didn't understand in the presentation, what happens to shadow mode when the car is FSD? The computer has 2 of Tesla's custom chips, for redundancy. They said both chips run the same code and compare answers to make sure they're functioning correctly.
Where does shadow mode fit in?
Last edited:
Imagine trusting humans to drive. It kills 30,000 people a year just in the USA!
Tesla also has radar for the front facing approx. 120 degrees. Amongst other overlaps, they also showed various aspects and discussions on other techniques, some dropped due to limitations and some areas extracted from visual data using the Tesla network. The impressive extraction of data from one single Telsa equipped car with the network showed a 3D model on the fly being generated, they also discuss real world conditions as based on visual cues and inputs from existing road/rules infrastructure etc.
Honestly do I feel like it's a slam dunk for Tesla? No.
Do I feel like the best team across many industries and tech is pushing their already market leading tech further and quicker than anyone else? Yes.
Do I feel they have a unique place in the market to deliver the best product now and long term? Yes.
Honestly do I feel like it's a slam dunk for Tesla? No.
Do I feel like the best team across many industries and tech is pushing their already market leading tech further and quicker than anyone else? Yes.
Do I feel they have a unique place in the market to deliver the best product now and long term? Yes.
So a 20 minute drive near Tesla's HQ and the driver had to put his hands on the wheel?
OP
OP
I think the fact that one of the rides needed a driver intervention proves it is the network running it(which is not feature complete yet), and not some pre baked smoke n mirrors programed route. Still trying to find more impressions.
Fairly sure Bezos has that title on lockdown for the foreseeable future
There is a certain charm and want to follow the potential success from the culmination of his ideas/companies.
- Solar panels on houses and buildings
- Solar arrays for large scale/commercial
- Batteries at home supplying appliances and devices while feeding excess energy back to infrastructure grids
- Large scale batteries for infrastructure
- Electric cars and commercial vehicles
- Cyclic and renewable by nature of design
In the last 10 years of his progress it's fucking impressive. Imagine if government and industry backed his methods for another 10 years where things could be...
In terms of dollar figures, may be. In terms of innovation and progress of humanity or affecting climate change issues etc? I'll take Elon thanks. There's more to measure a man or an idea with than just dollars.
yeah i was thinking more in terms of successful innovation than pure profit/money. I assume Jobs is really the only competitor to Musk in that regard
Since it seems this livestream has ended, and the product has been unveiled, please move this discussion here: https://www.resetera.com/threads/te...jectively-the-best-chip-in-the-world’.113049/
- Status