mikes_fsd
Banned
To be frank, he never answered the Third Row's question..
He mainly stated that changing lanes is not as big a challenge as making turns at city intersections. (which IMO is an accurate assessment)
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Future of FSD after "traffic light & stop sign response" release
- Thread starter diplomat33
- Start date
He mainly stated that changing lanes is not as big a challenge as making turns at city intersections. (which IMO is an accurate assessment)
By the way, here is a similar video on how to correctly drive in a roundabout that also gives some idea of what Tesla will need to solve in order to properly release the roundabout feature.
MrBadger
Badger out
willow_hiller
Well-Known Member
Was it this thread we were talking about the camera placement not being able to see oncoming traffic at intersections?
Green provided some crash footage of a Tesla (on humanpilot) that rolled through a stop sign and was T-boned from the right. The right b-pillar camera had a perfect view of the oncoming vehicle right up to the crash: green on Twitter
Green provided some crash footage of a Tesla (on humanpilot) that rolled through a stop sign and was T-boned from the right. The right b-pillar camera had a perfect view of the oncoming vehicle right up to the crash: green on Twitter
Last edited:
willow_hiller
Well-Known Member
Additionally, some more informative tweets from Elon. Elon Musk on Twitter
Third row: "can you add a section in the odometer to show how many miles the car has driven on autopilot"
Elon: "Ok"
Third row: "Elon, a bunch of tesla drivers outside california complain to me that some of the speed limits / map data on exit lanes etc aren’t always accurate and it’s a pain in the ass for them works well here in CA, plans to improve map data or read speed limit signs?"
Elon: "Yes, this is a high priority"
Everyday Astronaut: "Why can’t the camera read the speed limit signs? Is it because of the patent from mobile eye? It seems like that’s not a very novel thing and wise for the future of self driving cars"
Elon: "Coming soon"
Third row: "can you add a section in the odometer to show how many miles the car has driven on autopilot"
Elon: "Ok"
Third row: "Elon, a bunch of tesla drivers outside california complain to me that some of the speed limits / map data on exit lanes etc aren’t always accurate and it’s a pain in the ass for them works well here in CA, plans to improve map data or read speed limit signs?"
Elon: "Yes, this is a high priority"
Everyday Astronaut: "Why can’t the camera read the speed limit signs? Is it because of the patent from mobile eye? It seems like that’s not a very novel thing and wise for the future of self driving cars"
Elon: "Coming soon"
Thanks. This is good news.
Right now, AP does not automatically speed up when the speed limit increases. For example, if you are driving on a city street where the speed limit is 35 mph and you are leaving the city limits and the speed limit becomes 55 mph, AP stays at 35 mph and you have to manually increase the TACC speed to 55 to get the car to accelerate up to the new speed limit. This is something that AP should do automatically. Hopefully, when Tesla releases speed limit recognition, AP will be able to automatically adjust the speed to match the speed limit. This is an important FSD feature that needs to be done.
And I never believed that Mobileye's patent means that nobody is allowed to read speed limit signs with their cameras. That is too broad. The patent would only apply to a specific way of reading speed limit signs. So, I am sure Tesla has been working on reading speed limit signs with the cameras. In fact, that is a critical part of autonomous driving no matter what. I am pretty sure that Waymo, Cruise, Zoox, pretty much every company doing FSD also reads speed limit signs with cameras.
Soda Popinski
Member
Here's the text of the patent Claim #1:
Patent US20080137908A1 said:
Not particularly specific. This is the goal of patent attorneys, to get as broad coverage as possible.
Thanks. That is helpful. But other companies, including Tesla, can and will read speed limit signs with cameras without infringing on Mobileye's patent.
Soda Popinski
Member
There's a post on Reddit claiming "partitioning" means something specific, making the patent more narrow than I'd interpret.
That said, this patent is specific to "driver assistance function." So if Tesla ever gets driverless FSD working, it couldn't apply. One could even argue, since FSD is designed to eventually replace the driver, it's not really a driver assistance system. So the human is assisting the AI driver....
That said, this patent is specific to "driver assistance function." So if Tesla ever gets driverless FSD working, it couldn't apply. One could even argue, since FSD is designed to eventually replace the driver, it's not really a driver assistance system. So the human is assisting the AI driver....
Doesn't Waymo and Cruise use High Definition Maps? If so, perhaps the accuracy of speed is achieved using these maps over signage.
I don't know - just speculating.
Seems to me that it could be broken into at least three steps:
1) Straight through
2) Right turns
3) Left turns
1) Straight through
2) Right turns
3) Left turns
mikes_fsd
Banned
To say that is oversimplified would be an understatement.
- An intersection where the streets are not close to perpendicular to each other.
- A multi 3+ street intersection.
- An intersection that has at grade rail crossing at the intersection.
- Intersections that take you flow into another intersection (cannot block the perpendicular flow)
Yes, I was grossly over-simplifying with my own three bullet points ... if Tesla really is able to pull off all turns at all intersections in their next major release, kudos to them, but if they break it down into smaller steps, it will need to very simplistic so that they can remember.
Note that for any auto-turning to work the driver will need to have some kind of destination punched in.
I think the timeline is incredibly optimistic but it does summarize what I think Tesla will try to do with FSD. And since this thread is about the future of FSD, I think it is relevant.
Basically, Tesla will try to finish "turning at intersections", "roundabouts" and other stuff like speed limit signs this year to achieve "feature complete" where the car can complete a common commute on its own but with driver supervision and then after that is done, Tesla will try to refine the software with fleet feedback until driver disengagements go down to zero.
Basically, Tesla will try to finish "turning at intersections", "roundabouts" and other stuff like speed limit signs this year to achieve "feature complete" where the car can complete a common commute on its own but with driver supervision and then after that is done, Tesla will try to refine the software with fleet feedback until driver disengagements go down to zero.
powertoold
Active Member
Tesla's approach allows for big leaps in progress. I believe their progress after the traffic control update will surprise us. Within the traffic control update, we can see that AP is starting to detect different types of intersections, lane position, traffic light position, turn lanes vs straight lanes, etc. There's a lot of progress within the traffic control update that demonstrates proof of concept (that vision can reliably drive a car). Prior, we only had driver assist that was available with other mfgs.
Interesting - Then do you believe they pay MobileEye for use of their patented technology (e.g. Camera Vision to read speed limit signs)?
Can't wait to see how Tesla will work with at a major intersection that has 5 "points", where it is not just Straight, Left and Right, but Straight, Left, Right and "slightly right/left"
Google Maps
I imagine intersections like that will probably require driver intervention at least at first. Let's remember that Tesla is not promising FSD will work perfectly everywhere out of the gate. They are merely promising that this year or next year, they will have the pieces together that the car could do some complete drives without a disengagement. But there will be other driving scenarios and other types of intersections that the car won't be able to handle at first, that require more work from the AP team to solve.
I don't know. I've not read of Waymo paying Mobileye but that does not mean that they are not. It is possible that Waymo found a way to read signs with camera vision that does not violate the Mobileye patent.
Similar threads
