Definitely much improved in some areas but still falling well short in others. One step backwards for each one forwards is a clear indication of not coping with the scale of the complexities.
Nothing I've seen persuades me from the view that we will never have full self driving in private cars due to the diversity and complexities of their common usage. Clues in the video are having to take control in a car park to direct the car into the most likely lane where there could be vacant spaces, something that requires a human brain, previous experience and judgement. Being unable to escape a car park due to an unmarked exit that once again required uniquely human guesswork. The gambling brain in other words, not logic or AI.
And another very human aspect, courtesy. Stopping for a car reversing out of a bay but blocking the pedestrian waiting to cross, where I would have combined both with a friendly smile and wave across for the pedestrian. AI will never do that.
I also believe that the new Tesla approach of cameras and AI only and no Lidar or Radar is a big mistake that will bite them hard at times. Refraction, reflection and dazzle can be insoluble problems with only single plane polarisation available to deal with them if there's no Radar or Lidar. But bringing those two back returns to not being able to cope with the input complexities.
Answers are still a very long way away.
.
www.ncbi.nlm.nih.gov
