Don't get me wrong this is promising technology but I would be hesitant to rely on it to do too much.
I understand that argument but I am of the school of thought that the machine should not make decisions that can impact my safety unless it is an emergency. shaving off a few seconds on my commute is not what I would call a much of a benefit when there is a substantial possible downside.
The transition between research and deployment is an important period where things that recently became possible can be fully scrutinized. The goal of research is simply to make things possible. The rest is a long hard slog that not all tech makes it through.
In short, I really wouldn't worry about anything they're working on in a research facility, that's just so far away from shipping product figuring out which parts are more or less useful probably isn't worth doing at this point.
One tidbit: a number of stoplights in Palo Alto and Redwood City broadcast information about when they are going to change. A properly equipped car can receive that information and decide whether to proceed as usual or begin braking, even before there is a visible change. As a corollary, if the red light is about to turn green, the car can decide not to slow down.