Cops and Invisible Robbers: the Cost of Drunkenness
This work addresses a gap in graph pursuit game theory by introducing and analyzing the invisible robber variant, offering theoretical insights but is incremental as it builds on existing Cops and Robber frameworks.
The paper tackles the Cops and Robber game with an invisible robber, analyzing the invisible Cost of Drunkenness (iCOD) as the ratio of capture times between adversarial and drunk robber variants. It provides exact asymptotic values for specific graph families like d-regular trees, bounds for grids, and shows iCOD can range arbitrarily from 2 to infinity.
We examine a version of the Cops and Robber (CR) game in which the robber is invisible, i.e., the cops do not know his location until they capture him. Apparently this game (CiR) has received little attention in the CR literature. We examine two variants: in the first the robber is adversarial (he actively tries to avoid capture); in the second he is drunk (he performs a random walk). Our goal in this paper is to study the invisible Cost of Drunkenness (iCOD), which is defined as the ratio ct_i(G)/dct_i(G), with ct_i(G) and dct_i(G) being the expected capture times in the adversarial and drunk CiR variants, respectively. We show that these capture times are well defined, using game theory for the adversarial case and partially observable Markov decision processes (POMDP) for the drunk case. We give exact asymptotic values of iCOD for several special graph families such as $d$-regular trees, give some bounds for grids, and provide general upper and lower bounds for general classes of graphs. We also give an infinite family of graphs showing that iCOD can be arbitrarily close to any value in [2,infinty). Finally, we briefly examine one more CiR variant, in which the robber is invisible and "infinitely fast"; we argue that this variant is significantly different from the Graph Search game, despite several similarities between the two games.