Appears in collection : Frontiers of reconnaissability / Frontières de la reconnaissabilité

We show that two-player stochastic games with perfect-information and shift-invariant submixing payoff functions are half-positional, i.e. in these games the maximizer has a positional optimal strategy. This extension of our previous result for one-player games relies on an interesting existence result about the existence of epsilon-subgame-perfect strategies.