with the release on software for hardware 2.0 we have some more 'data points'
thus there is some more food to think on topic question.
1. those frequent 'red cars' false positives with hardware2 are due to lack of adequate 3D reasoning, while info from Tesla indicates radar has some 3D capabilities, it looks like essentially it is 2D radar with 2.5D capabilities just due to software processing, thus so lot of false positives are needed to map. But any new road, or road with significant upgrades could potentially lead to accidents - so such situation could not end with 4/5 self driving capabilities because just there always be situations, when active driver intervention is needed.
3D radar of lidar or any other tech which does not need to map roads with humans to start reasoning about unknown surroundings is really needed, also those difficulties to detect stationary cars should not exist at all for any true level 4/5 self driving car, and sadly they are here for now.
2. with above observations it looks there is no way that current hardware technology could attain level 5 self driving, possibly not even level 4 ( though level 3+ is certainly will be attainted ),
for level 4/5 there is clear need for more sensors and exactly those sensors which perceive true 3D
and we can try to make new estimations for advised level4/5 self driving capabilities:
cheap lidars ( and the likes ) start to appear, but could be two years away for market availability, there are also seems difficulties with software ( because it took too much time from Google even when they disregarded costs of previous generation of lidars )
radically better 3D radar takes time to develop - so we could assume that it could appear no faster than past 2020
but at the same time by the end of the year it will be crystal clear, that Tesla has solid software and hardware for level 3+ self driving car but no more, there will be attempt to make some modifications - either use more precise gps and annotated maps as assistance to bring software/hardware to level 4 in some areas or there will be rush to install lidars or the likes, which can figure out 3D much better, than current hardware, but most probably - lidars will be skipped again, and then level 4 self driving capabilities will be reached by end of 2018 but only past 2020 we could expect some true level 5 autonomous driving with those mentioned true 3D sensing hardware.