As a FSD user, I personally find that it works quite well on highways. I was traveling this weekend and it adeptly managed the entire route from our home through 40 miles of highway to the entrance to short term parking at the airport we were flying out of, then back from the airport to our home. The last few hundred feet e.g. parking structures it does not handle well, however. On highways, it lane changes to pass slower traffic, reacts to abrupt slowdowns, can execute an aggressive merge while being courteous with other drivers and generally shows uncannily human behavior. V12 with the fully trained stack is quite impressive.
I am far less comfortable using it on city streets - autonomous turns, even unprotect left turns are handled amazingly well, however lower speed lane merges are frequently botched, particularly with indicated yields. It works, but the cognitive load required of me to supervise it is greater than that I would expend by driving myself.
China is a good use case for them to pursue - many modern roads, high speeds, wildly varying driver experience and excellent maps.
There is no question in my mind that this is the future. I've done a few dozen Waymo rides (fully driver-less) and other than the last 100' not always being elegantly handled, every ride has been flawless.
A key technical challenge in this space is the reliance on maps which is what this deal is about. There are vastly different schools of thought here, from an engineering perspective. One is that the availability of high quality, high resolution maps with indicated features (stop signs, merges, crosswalks etc.) dramatically increase performance... obviously. But, the autonomous system cannot assume that its stored maps are accurate--maps go out of date. So, by choosing to utilize maps, you inherit the engineering problem of having to assess in operation whether the map is accurate and, if not, react accordingly. Without maps, perception of the world is so much more critical, but if you do it properly then the vehicle should enjoy a wider operating envelope including in areas where there are no maps. So... it's an interesting problem. I can think of all sorts of interesting data access problems... where is the mapping data stored? Does Tesla USA have access to it?
