People who bend the rules like that are a problem to be fixed, not one to be accomodated. I would train the self driving cars to wait until the appropriate conditions are met, and if that means until that person stops waving and continues appropriately, all the better. Road rules are as much about predictability as they are about safety. As long as you can predict another road user will behave a certain way, you can ensure that you can navigate a situation safely without further communication. The moment they start to bend the rules, they introduce uncertainty into the system, which increases the need for communication and the likelihood of mistakes.