This chapter demonstrates an application of agent-based selection dynamics to the traffic assignment problem. We introduce an evolutionary dynamic approach that acquires payoff data from multi-agent reinforcement learning to enable an adaptive optimization of traffic assignment, provided that classical theories of traffic user equilibrium pose the problem as one of global optimization. We then show how these data can be employed to define the conditions for evolutionary stability and Nash equilibria. The validity of this method is demonstrated by studies in traffic network

This paper presents a relationship between evolutionary game dynamics and distributed recency-weighted Monte Carlo learning. After reviewing some existing theories of replicator dynamics and agent-based Monte Carlo learning, we provide proofs of the formulation-level equivalence between these two models. The relationship will be revealed not only from a theoretical viewpoint, but also by computational simulations of the models. As a consequence, macro dynamic patterns generated by distributed micro-decisions can be explained by parameters defined at an individual level. In particular, given the equivalent formulations, we investigate how the rate of agents’ recency weighting in learning affects the emergent evolutionary game dynamic patterns. An increase in this rate negatively affects the inertia, making the evolutionary stability condition more strict, and positively affecting the evolutionary speed toward equilibrium.