I'VE been thinking about this for a while - since the auto trans is documented as "leaning" the way you drive and adjusting accordingly, is it possible that some of the issues people are seeing could be due to the drive patterns exhibited during test drives? I'd love to hear thoughts on the theory from those familiar with the technology.
I wouldn't think so, unless it tries to learn really fast. For previous learning transmissions, it usually takes a few tanks of gas in my experience.
I think at least in part, the problem is one of perception. CAFE regs are demanding more efficiency. AWD vehicles are getting hit first, and the people driving them are beta testers right now for several brands. They have new engineering solutions to the problem, and just aren't the same as what we are used to. Some of what they do strikes us as weird.
For example, my co-worker just bought a 2015 forrester. It's a CVT. It does the CVT thing of driving like a boat where you hit the optimal RPM and goose it or not, it doesn't move out of that band much. That doesn't bother him, but what he hates (in large part because it turns exiting his driveway into a death defying act every morning), it the fact that the forward gears are CVT, but reverse is actually a fixed ratio singe gear interface. Switching between them takes about 1-1.5 seconds to happen according to him. That leaves his butt hanging in the breeze every morning when he backs out of his driveway and make the parking lot at work fun too.
The Zf 9 speed is complex. Not all the shifts are similar, it seems to figure in all sorts of skip shifting patterns into it's logic, and I suspect the learning transmission thing (which isn't new), changes from adjusting to achieve a single goal to having to develop an overall shifting strategy for a given driver. That's a much, much harder task. Then of course you graft on AWD, and in the case of jeep a much more complicated AWD offering than most from a logic standpoint, and then add on things like stability control, then add on things like hill descent, and I suspect you have a lot of assumptions going into this thing that make a much bigger difference to how it feels than your typical fuzzy logic learning transmission software of the past.