Anyone looking at the poor quality video (and the actual video was better) would instantly recognize a woman walking a bike. The object recognition is not ready for the real world and probably won't be for another decade or two.
Also, it seems that a collision with anything should have been grounds for an avoidance and braking. I cannot guess why software engineers would want a system under development to not activate emergency procedures in the event of an eminent collision. The concept of avoiding erratic driving does not really make sense, since the software would in theory make the best of the situation.
Or else as you post, the system is not ready for prime time. In which case why would it be tested on actual roads? With actual pedestrians with bicycles?
That woman was also drunk (and an ex-con for numerous thefts and drug crimes) who did not once look to see if a car was coming, despite walking across a busy road away from a crosswalk and away from overhead lights.
And in SIX SECONDS did nothing to avoid killing the woman. The driver was busy was readouts instead of the road. Park these cars!
Nothing about these systems is ready for prime time. It’s being lied about and over sold. This tech is DECADES away. These vehicles should not be on public streets.
There are literally thousands of edge cases they have no clue how to solve.
They don’t consider these systems safe enough to self drive trains or semis on the interstates which are all MUCH SIMPLER problems, yet these things are on urban streets, its unfathomable.
Uber committed murder, and the local, state and federal governments that let these things on the road are culpable.