InsanePath
Member
Glad I'm not the only one experiencing not being in the center of the lane. I just thought maybe it was just me lol.
What "innovation" are you talking about? Moving the car away from the semis for no reason is the innovation I can live without. Not sure why anyone would welcome a behavior like that but should be an optional setting if anything.
It is not limited only to the leftmost lane. I don't have an issue with code to allow for lane splitting, but I think part of that can be integrated into the code that already moves over if something encroaches into your lane.One observation that I am still validating is that it only occurs in the left most lane. That would point to a very specific use case that has been discussed for awhile now. In California, there has been tremendous complaints about Tesla vehicles with AP engaged does not move over for motorcycles. They primarily split vehicles between the leftmost and adjacent lanes. This especially is a problem with large trucks. I can appreciate it isn't valid in all states, but will be appreciated where lane splitting is legal. In Texas, there is no law against it and does happen, but not as common as in CA. Also, as long as I have been following this group, many drivers have complained how close the vehicles get to semi trucks. Personally, I think that was more of a perception issue than an actual issue. But, it seems to address that as well. So, innovation...yes. As a software design lead, the innovation is very difficult to execute and will get better over time.
It is not limited only to the leftmost lane.
I might be guilty of enforcing some of this behaviour as I used feel my car was too close to trucks when passing them on the high-way and I disengaged the NoA by turning left to clear more space between me and the truck.
This is then interpreted as an "error" in the autopilot by tesla and the machine learning things correct for it.
Now after a while I trust the software a bit more and stopped doing this as I figured the car would not crash anyway, but if many people did what I did it might be used to train the network.
That’s not how the machine learning works. Tesla uses machine learning to identify objects, but *not* to define driving rules. Those rules are actually pretty basic and, as such, well-suited to being hard-coded in the software,