Starting a new thread on the wonderful Stop Sign / Light feature rolling out to cars with FSD.
Are we teaching ours cars how to stop at Stop Signs and Stop Lights?
I believe we are based on these observations:
- The first time the car encounters a Stop Sign/Light it does so using the only the camera. This is clear because the messages on the screen indicating the Sign/Light is coming, do not appear until the Sign/Light is actually visible.
- Once the car has encountered an intersection for the first? time, then the car "remembers" where that intersection was. It has been trained.
- The second time it encounters the same intersection, it is aware way before the Sign/Light is visible (obscured by trees, curves, hill, etc.). This is clear because the message about the upcoming intersection appear on the screen way before it is visible.
- Once an intersection is "remembered" in this fashion, by any of us in the fleet, all of our cars subsequently know about that intersection.
After only a week of driving around my town now all the Signs/Lights have been "remembered". Messages appear before the intersection is visible.
This is the first time I am aware of Tesla's Autopilot using information from a map to drive the car.
Tesla Autopilot does not use maps to drive. It drives solely based on what it sees with its eight cameras and one radar. We know from Autonomy Day that Elon is against maps (however up to date or detailed) providing anything other tertiary information. Like planning a route for navigation. This may be one of the first verifiable uses of map information beyond that.
GM's Super Cruise and Waymo's autopilot systems, on the other hand, very map dependent. In fact, they are geo-fenced. They will only operate on highways they have maps for.
Tesla's Autopilot will enable anywhere it can identify a lane line.
If true, this Stop Sign/Light feature is a very good example of fleet learning. We are training the neural net.
Has anyone else experienced this?
Are we teaching ours cars how to stop at Stop Signs and Stop Lights?
I believe we are based on these observations:
- The first time the car encounters a Stop Sign/Light it does so using the only the camera. This is clear because the messages on the screen indicating the Sign/Light is coming, do not appear until the Sign/Light is actually visible.
- Once the car has encountered an intersection for the first? time, then the car "remembers" where that intersection was. It has been trained.
- The second time it encounters the same intersection, it is aware way before the Sign/Light is visible (obscured by trees, curves, hill, etc.). This is clear because the message about the upcoming intersection appear on the screen way before it is visible.
- Once an intersection is "remembered" in this fashion, by any of us in the fleet, all of our cars subsequently know about that intersection.
After only a week of driving around my town now all the Signs/Lights have been "remembered". Messages appear before the intersection is visible.
This is the first time I am aware of Tesla's Autopilot using information from a map to drive the car.
Tesla Autopilot does not use maps to drive. It drives solely based on what it sees with its eight cameras and one radar. We know from Autonomy Day that Elon is against maps (however up to date or detailed) providing anything other tertiary information. Like planning a route for navigation. This may be one of the first verifiable uses of map information beyond that.
GM's Super Cruise and Waymo's autopilot systems, on the other hand, very map dependent. In fact, they are geo-fenced. They will only operate on highways they have maps for.
Tesla's Autopilot will enable anywhere it can identify a lane line.
If true, this Stop Sign/Light feature is a very good example of fleet learning. We are training the neural net.
Has anyone else experienced this?
