As I understand, with PoS the selected validator is determined within predictable time windows. Does this mean that uncles will be gone for PoS?
As it stands, by my reading of it, there is something akin to uncles in the proposed PoS system, referred to as "dunkles". See this:
The second strategy is to simply punish validators for creating blocks on the wrong chain. That is, if there are two competing chains, A and B, then if a validator creates a block on B, they get a reward of +R on B, but the block header can be included into A (in Casper this is called a "dunkle")
And from the (possibly superceded) mauve paper by Vitalik Buterin:
The purpose of the "dunkle" mechanism is to solve the "nothing at stake" problem in proof of stake, where if there are no penalties but only rewards then validators are incentivized to try to make blocks on top of every possible chain.
If a validator bets on a block that doesn't make it into the chain (a dunkle), then they will suffer a penalty when the dunkle's header is included in a future valid block (with a reward paid to the validator that includes it).