diplomat33
Average guy who loves autonomous vehicles
Mobileye SuperVision "L2 door to door" to go to beta testers in China on the Zeekr001:
-
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
I read or saw a video on China's advancement on FSD, and heard that one of the lead techs left Tesla and took the code with him to China, and there was a lawsuit against him from Tesla which was settled. I wonder if all the China companies coming out with FSD started with some basis of Tesla's code and then forked it to their platform...
diplomat33
Average guy who loves autonomous vehicles
The stuff he allegedly took was AP, before FSD was really a thing.
Tesla settles lawsuit against engineer who it claims stole Autopilot source code for Chinese competitor
XPeng is using a solution that is based on NVidia Drive OS, so there really isn't much need for going on their own. There's a bunch ready-made solutions available so OEMs now have a much easier time (they only need to add their own branding and tweaking for their local market).
Nvidia Drive - BEYOND HORSEPOWER: XPENG P7 SHOWS WHAT’S NEXT FOR INTELLIGENT TRANSPORTATION
Faster than I thought. They just recently pushed the highway feature and now about to start testing the city streets navigation. So now there is the Xpeng City NGP and Mobileye Supervision and next GM Ultra Cruise coming next year.
IMO, it's ridiculous to try and develop an autonomous system that doesn't rely on maps. What percentage of the roads are unmapped? 1%? Less? Surely people can still be depended on to drive those unmapped roads until AV systems have conquered the rest of the 99%. The one thing I know computers are better than humans at is aggregating and processing lots of information quickly. Why limit in the input data for them to function?
diplomat33
Average guy who loves autonomous vehicles
Argo robotaxis officially available to the public now on Lyft in Austin. They will be for the same price as a normal Lyft ride. There will be a safety driver and a specialist in the front passenger seat to monitor the ride at first.
techcrunch.com
Argo AI's robotaxis officially available via Lyft in Austin | TechCrunch
Argo AI and Lyft have launched a commercial robotaxi service in Austin, however safety operators will still be present in the vehicles.
techcrunch.com
Of course we need maps to navigate, but these do not have to be accurate to an inch. We have navigation maps already, such as Google Maps. But what about highly detailed (HD) maps?
The real question is, how many roads are or can be correctly mapped in high resolution? I think, less than 99%, because the real world frequently changes. So each autonomous car has to check for deviations from the map all the time, because it can never be sure.
Difficult to say what that means. It may mean that highly detailed mapping is useless altogether, once autonomous cars are good enough to see and understand reality sufficiently well.
Certain parts of mapping may always be useful, for example, traffic signs. Since they are sometimes obscured, it is useful to know what other cars previously saw. Potholes are another, similar issue, because they cannot always be recognized in time. When it is wet, every puddle can hide a pothole underneath. It is probably also useful to know where marked lanes begin and end and where they lead to.
But this does not mean that total, complete, highly detailed mapping is needed everywhere. Perhaps it is quite sufficient to map only problematic spots. Perhaps it is sufficient to mark 95% of the roads as normal, unsurprising, and only map the difficult 5% in detail.
diplomat33
Average guy who loves autonomous vehicles
Intel files IPO plans for Mobileye:
www.barrons.com
Intel Files IPO Plans for Its Mobileye Autonomous Driving Unit
Intel acquired Mobileye for $15.7 billion in 2017. The company could potentially fetch twice that valuation in an upcoming IPO.
www.barrons.com
petit_bateau
Active Member
Just like VW putting Porsche to an IPO at this point. Because they are dead men walking.Intel files IPO plans for Mobileye:
Intel Files IPO Plans for Its Mobileye Autonomous Driving Unit
Intel acquired Mobileye for $15.7 billion in 2017. The company could potentially fetch twice that valuation in an upcoming IPO.
diplomat33
Average guy who loves autonomous vehicles
Who are dead men walking? Definitely not Mobileye. Mobileye has the best FSD approach for scaling both robotaxis and self-driving personal cars.
I was specifically thinking about things like
lane selection, speed limits, school zones, intersections, and such. Not simply navigation and not HD maps down to inches. It is specifically for things like lane selection that Tesla is getting rid of map
Information in favor of reading lines from vision. And lane selection has gotten HORRIBLE! It really burned me to see the team up there on AI day touting their new lane selection NNs and how they went from vision models to language models for lane selection and how cool it was. Makes me wonder if any of these guys actually drive a Tesla with FSDb. Of course, all their commutes/routes (as well as all of Elon’s) are probably in the confirmation/testing data they talked about, so every release probably works just fine for them.
diplomat33
Average guy who loves autonomous vehicles
Part 2 of Brad Templeton's "why you don't have a self-driving car yet?" video. It focuses on the social/business/regulatory challenges to deploying self-driving cars.
0:00 - Intro
0:32 - Road citizenship
3:22 - You live in the wrong town
6:55 - PuDo
8:20 - Business model
9:30 - APP
9:55 - Making it too safe
11:20 - Public and government acceptance
13:40 - Non blockers
15:02 - So, when?
0:00 - Intro
0:32 - Road citizenship
3:22 - You live in the wrong town
6:55 - PuDo
8:20 - Business model
9:30 - APP
9:55 - Making it too safe
11:20 - Public and government acceptance
13:40 - Non blockers
15:02 - So, when?
I thought they shipped the opposite in the recent FSD update, it's in the release notes. They now feed in coarse map data (lane counts, turn restrictions, etc) as inputs to the network so it likely biases toward the map data within the weights of the network. Seems like a good approach as opposed to trying to hand-code some function to decide to use maps or not use maps. Then if vision is unambiguous that the map data is wrong, it will hold little weight on the outputs.
Well that specifically is being addressed in FSD version 10.69.2.3.
“Added a new "deep lane guidance" module to the Vector Lanes neural network which fuses features extracted from the video streams with coarse map data, i.e. lane counts and lane connectivities. This architecture achieves a 44% lower error rate on lane topology compared to the previous model, enabling smoother control before lanes and their connectivities becomes visually apparent.”
Which means they have a new feature that has more info on lane configurations which should allow better lane choices. They just have their own system utilizing their massive video database to extract lane information instead of relying on static maps. I assume that when lane configurations change the new car videos would show that so the database could be updated quickly.
Grocery delivery in Canada:
www.cp24.com
"Loblaw Cos. Ltd and autonomous middle mile logistics company Gatik have teamed up to launch Canada’s first fully driverless grocery delivery truck.
...
As of today, Loblaw told CTV News Toronto in an email it has five trucks on the road. One of those vehicles is operating fully autonomously, while the other four have a safety driver on board."
Loblaw's driverless grocery delivery trucks hit the streets of Toronto
Shopping for groceries in the Greater Toronto Area just got a little more futuristic.
www.cp24.com
L2 door to door would be great. Hands free driving. 95 percent of my driving are 2 lane highways and the dirt road I live on
When you call for a Waymo, Cruise or Argo does it go through a dispatch center then they send the car that is closest to you? Or does the request go through a car and the car knows exactly where you are and knows where the car needs to go with zero human intervention. Will it be possible to have the car recognize voice commands and take you to your destination based on where you tell it to go and adjust price accordingly. Say I'm going to the Supermarket but once I am in the car I decide I want to go to the Walmart Supercenter. Will I be able to say take me to Walmart in this town?
Similar threads
- Article
