Solution Concepts for the Simulation of Household-Level Joint Decision Making in Multi-Agent Travel Simulation Tools

In the recent years, there have been a growing interest in understanding and forecasting joint travel-related decisions, that is, decisions taken by several individuals together, including binding agreements. Such forecasting would first allow to predict the impact of policies aiming at influencing this kind of behavior (for instance policies aimed at increasing car occupancy), but also improve the forecasts in general, by taking into account the effect of spatial dispersion of social contacts when choosing a joint leisure location, for instance. A large number of attempts at simulating the state of transport systems have had a game theoretic view: individuals are seen as agents getting a utility from their travel decisions, this utility depending on the decisions of others (mainly via congestion). Game theory is aimed at defining and studying solution concepts for such situations, that is, ways to predict probable outcomes of such games. Most of the research in travel behavior forecasting relied on the equilibrium family of solution concepts. In this setting, individuals are seen as selfish agents competing for limited joint (capacity) resources. Another field where game-theoretic concepts have had important impact is in the field of co- evolutionary computation. An important stream of literature in this field in particular insists on the importance of the game-theoretic solution concept explicitly or implicitly underlying the search process, which will favor one solution or the other. Equilibrium is not the only solution concept from game theory, and its applicability in the case of decisions taken amongst emotionally related individuals, such as household members, is dubious. In particular, the possibility of realizing binding agreements is excluded from such formulation. This paper uses a co-evolutionary algorithm, built using the MATSim software framework, to investigate the usage of two different solution concepts for the problem of predicting intra- household joint travel: (a) a “Group Utility” concept, classical in the research on household decision-making, and (b) an “Absence of Blocking Coalition” concept, which allows to represent selfish but coordinating players with arbitrary social network topologies. The implementations of those solution concepts in MATSim are used on a scenario for the Zurich area, to reveal their strengths and weaknesses.