I wonder if Tesla looks at the video feed from the camera. I have a stretch of road near my home that is very lousy for AP. I run it there to see if it is even possible for them to solve. So far it does seem less jerky. For this road I think they would need to rely more on GPS than the camera.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
How quickly is autopilot "learning" about its trouble spots? Anybody experience it?
- Thread starter calisnow
- Start date
-
- Tags
- Autopilot Tesla Inc.
Here is a simple way:
1. Car is on autopilot
2. User performs (extreme?) action
3. Autopilot tries to figure what user did
4. Autopilot stores GPS position and some extra data along with user action end result
5. Voila - autopilot has learned. No download needed.
The car is constantly sending telemetry data back to Tesla and Tesla is constantly updating AutoPilots algorithms based on new telemetry data. This is the key component to AutoPilot's system that is leaps and bounds beyond what other cars are doing.
Tesla is a software company that happens to make amazing cars and batteries. Other companies are 5-10 years behind Tesla when it comes to software.
Tesla is a software company that happens to make amazing cars and batteries. Other companies are 5-10 years behind Tesla when it comes to software.
mknox
Well-Known Member
Are these tweaks pushed down to cars continuously, or would they be aggregated and built in to the next firmware release? If the former, would someone buying a brand new car have access to these improvements, or would the car have to learn on it's own over time?
My guess, based on how software development/release is usually done, is that the firmware updates are separate from any auto-steering tweaks, and possibly other such data feeds. My assumption is that the firmware has a set of algorithms for dealing with the data, and the data keeps being updated so it's available to the firmware, but firmware updates are needed to change the algorithms. Otherwise the number and frequency of the releases would be a problem and counterproductive since only the data has changed.
As a software developer my guess is that there are autopilot "hints" downloaded along with the nav maps/data. So the code is the same but it consults these hints based on location and direction of travel when making decisions. The hints could be updated constantly based on driving data.
If true I'd be curious how many of the hints are able to be generated automatically through bulk data processing and how many of the hints would need to be created by a human reviewing video/sensor data.
If true I'd be curious how many of the hints are able to be generated automatically through bulk data processing and how many of the hints would need to be created by a human reviewing video/sensor data.
This is exactly how I would do it - autopilot data would be treated as additional metadata for each mapping tile, with the live system in the car referring back to the metadata when there was a low certainty from the real-time data, i.e. an unpainted section of the road.
What would be very interesting is if they are also collecting junction data, so, for example, if a large number of cars stop at a specific location, the metadata would reflect the junction and initiate a "hands on" warning as the car approaches it.
For this to work in a dynamic environment, the metadata would need to "age", to cope with temporary stops like road works.
http://www.teslamotors.com/blog/your-autopilot-has-arrived
I guess this means the system is improving itself?
Todd Burch
14-Year Member
I've been looking for that (because for me that's the scarier scenario...that you get comfortable in a situation because the car's been handling it fine but later something unexpected happens). Surprisingly, I can't think of one case where that's happened--at least for me.
Vger
Active Member
My guess is Autopilot is a system that can work all on its own, but it is assisted by metadata based on GPS and just uses automated machine learning. Stuff that requires a firmware update is how the system uses and interprets and reacts to the metadata.
The metadata can be simple values that help give the autopilot weight on how to make its decisions. These values are created by when a majority of Teslas make the same decision in a hold or disengage situation. For example, when 45 of 50 Tesla trips on a highway exit don't take the exit, the metadata tells the autopilot at this GPS location that the left lane marker should be the highest priority over following a car or using a right lane marker. I don't think Tesla requires any manual review whatsoever for this situation--they probably manual review things that Autopilot can never get right, for example turn curve speeds (which needs an update to fix).
This metadata wouldn't be much different than a navigation map having street numbers loaded in, business, names, etc.--even if Tesla has few dozen variables for every section of street, this data would only need kb worth of data to stream while on the move.
I'm guessing it's sort of like a video game replay file, where you can watch a 1.5 hour long 4k hd "video" with a file measuring just a few dozen mb in size. The file just contains what commands were issued during the game, and the program just recreates what happened exactly without actually recording a video.
My opinion anyways!
The metadata can be simple values that help give the autopilot weight on how to make its decisions. These values are created by when a majority of Teslas make the same decision in a hold or disengage situation. For example, when 45 of 50 Tesla trips on a highway exit don't take the exit, the metadata tells the autopilot at this GPS location that the left lane marker should be the highest priority over following a car or using a right lane marker. I don't think Tesla requires any manual review whatsoever for this situation--they probably manual review things that Autopilot can never get right, for example turn curve speeds (which needs an update to fix).
This metadata wouldn't be much different than a navigation map having street numbers loaded in, business, names, etc.--even if Tesla has few dozen variables for every section of street, this data would only need kb worth of data to stream while on the move.
I'm guessing it's sort of like a video game replay file, where you can watch a 1.5 hour long 4k hd "video" with a file measuring just a few dozen mb in size. The file just contains what commands were issued during the game, and the program just recreates what happened exactly without actually recording a video.
My opinion anyways!
My 2016 BMW has very limited cloud connectivity. It shows weather, traffic, and album covers on mp3 music. It can schedule service appointments and notify the dealer of problems. -- Doug
Gotta love infinite!
That is only the infotainment system. Just one component of the car.
Similar threads
- Article
- Article
