This paper aims to understand how to guide travelers' routing behavior toward a system optimum, using a bilevel game in a multi-agent traffic environment. With the goal to optimize some systematic objective (e.g., overall traffic condition) of city planners, we formulate a Stackelberg game with the upper level as the planner and the lower level as a multi-agent Markov game in which each rational and selfish traveler aims to minimize her travel cost. We employ a Bayesian optimization method on the upper level to solve for optimal controls of city planners and a mean field multi-agent deep Q learning approach to solve for optimal route choices of travelers on the lower level. We demonstrate the effect of two administrative measures, namely tolling and signal control, on the behavior of travelers on the Braess network and a large-sized real-world road network, respectively. Braess paradox, which is usually defined in static user equilibrium, is also defined and discovered in the context of the multi-agent Markov game.
Bayesian Optimization for Multi-Agent Routing in Markov Games
08.10.2022
1283131 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Drone Delivery Multi-Agent Routing Optimization
AIAA | 2020
|DRONE DELIVERY MULTI-AGENT ROUTING OPTIMIZATION
TIBKAT | 2020
|