My guess is as follows - even if GPS data is limited to a 3.5 meter lateral "box" - that is enough to give the car additional confidence as to which lane it is in when combined with other signals such as surrounding cars and poor, but not "completely absent" lane markings.
You would not need GPS data to tell you within inches where you are for the car to be able to use that data to keep it in its lane. You would simply need to know that cars tend to stay centered in a lane, and so if the GPS data is reading within plus or minus X feet then it is most likely the case that the car is in the number 2 lane, for example. Then that little bit of confidence can be combined with a little bit more confidence given by faded, but perhaps not "non existent" lane markings to add up to more confidence, etc. So in this way current GPS could in fact at least increase the likelihood of the car staying in its lane in the absence of clear guidance from the lane markings - especially if there are nearby cars for the Model S to add to its confidence picture.
If this is the case then it would make sense that over time, thousands of Teslas tracing "I am plus or minus X meters going in a straight line" combined with the knowledge that cars tend to travel in the center of a lane, more or less - would allow Tesla to use the statistical average of that imprecise GPS data to build high precision maps of a highways individual lanes.
What Tesla has not told us of course is just how much data they need from the user base on a given stretch of road to build out those maps. 10 passes? 100 passes? More?
You would not need GPS data to tell you within inches where you are for the car to be able to use that data to keep it in its lane. You would simply need to know that cars tend to stay centered in a lane, and so if the GPS data is reading within plus or minus X feet then it is most likely the case that the car is in the number 2 lane, for example. Then that little bit of confidence can be combined with a little bit more confidence given by faded, but perhaps not "non existent" lane markings to add up to more confidence, etc. So in this way current GPS could in fact at least increase the likelihood of the car staying in its lane in the absence of clear guidance from the lane markings - especially if there are nearby cars for the Model S to add to its confidence picture.
If this is the case then it would make sense that over time, thousands of Teslas tracing "I am plus or minus X meters going in a straight line" combined with the knowledge that cars tend to travel in the center of a lane, more or less - would allow Tesla to use the statistical average of that imprecise GPS data to build high precision maps of a highways individual lanes.
What Tesla has not told us of course is just how much data they need from the user base on a given stretch of road to build out those maps. 10 passes? 100 passes? More?