When two cars are facing each other, there is a lot of human interaction and understanding about who should go first, who should pull off into a parking spot, or perhaps who should reverse into the intersection to let the other pass.
This is one of thousands of edge-cases that are going to be extremely challenging to automate. I'm not sure it's even possible without some understanding of the counterparty, their intentions and body language.
One problem with developing these technologies in California is that California has very wide, very regular roads. High visibility and dry conditions are the norm. There are few uncontrolled intersections (in Seattle they are the rule, not an exception). I think California's history as a motorist culture has conversely made it one of the easiest places to automate.