The feature is available in more than just their sports car, you know. It's also in their sedan, their crossover SUV, etc.
I don't think it has anything to do with conflicting niches so much as having to spend time reinventing the wheel. After they lost MobileEye, they had to spend time redoing what MobileEye provided, plus everything they were planning to do going forward.
IMO, the big open question is whether the current AutoSteer tech is actually the basis for their self-driving tech, or just a temporary band-aid intended to replace the AP1 MobileEye functionality in the interim until their self-driving tech is ready.
Right now, I've seen the following problems (consistently) with AutoSteer:
- If you put on your turn signal, the Tesla either immediately changes lanes without giving enough time to warn other drivers or it does nothing at all, and as far as I can tell, there's no rhyme or reason to which of those two things happens.
- Sometimes when you put on the turn signal it tries to change lanes into a lane that is occupied by another vehicle. It never waits for a vehicle to get past you.
- On curves, it steers way too late (a full second after a good driver would do so), then turns the wheel too far, ends up veering towards the other lane edge, then swerves back and forth drunkenly for ten or fifteen seconds.
- On some curves, this results in the car leaving the lane entirely.
- When cars are in the adjacent lane, it does not favor the other side of the lane as it should.
- When there's a concrete barrier right next to the lane, it does not favor the other side of the lane as it should (and in many of those cases where it steered too late, I had to seize control to keep it from wrecking).
- It makes no attempt at maintaining a constant turning radius (which is the very first lesson that new drivers typically learn in driver's ed class)
- It usually fails to detect pedestrians and cyclists (even when they're crossing the road right in front of it).
- It doesn't respect traffic lights or stop signs.
- IMO, it doesn't brake soon enough when cars cut into the lane in front of you.
All in all, it isn't a beta so much as a pre-alpha. It is good enough for some freeways (the ones without significant turns), but it has trouble even on some four-lane, divided highways in the greater Bay Area, where presumably Tesla should have copious amounts of training data. I would have no faith in it on arbitrary roads. It isn't the edge cases that are wrong, but rather that the base case behavior is barely even adequate. It feels like they trained their model with drunk drivers and 15-year-old student drivers.
So I really hope that AutoSteer is a temporary replacement for MobileEye, and that the reason it isn't better is that it is getting only minimal maintenance. If that's not the case... we could be waiting a while.