Reaching Your Goal Optimally by Playing at Random

Shortest-path games are two-player zero-sum games played on a graph equipped with integer weights. One player, that we call Min, wants to reach a target set of states while minimising the total weight, and the other one has an antagonistic objective. This combination of a qualitative reachability objective and a quantitative total-payoff objective is one of the simplest setting where Min needs memory (pseudo-polynomial in the weights) to play optimally. In this article, we aim at studying a tradeoff allowing Min to play at random, but using no memory. We show that Min can achieve the same optimal value in both cases. In particular, we compute a randomised memoryless $\varepsilon$-optimal strategy when it exists, where probabilities are parametrised by $\varepsilon$. We then characterise, and decide in polynomial time, the class of games admitting an optimal randomised memoryless strategy.