Strategic Infrastructure Design via Multi-Agent Congestion Games with Joint Placement and Pricing

A team of researchers has introduced a novel AI framework that tackles the complex challenge of planning infrastructure like EV charging networks. The work, led by Niloofar Aminikalibar, Farzaneh Farhadi, and Maria Chli, formalizes the problem as a bi-level optimization model. The 'upper level' represents a central planner making decisions on where to place resources and how to price them, while the 'lower level' models how self-interested agents—like EV drivers and regular commuters—respond to those decisions, using a concept from game theory called non-atomic congestion games.

To solve this computationally difficult (NP-hard) problem, the team developed a double-layer approximation framework named ABO-MPN. This method intelligently decouples different agent types and applies integer adjustment and rounding to find high-impact solutions. In experiments on benchmark networks, their model demonstrated a significant impact, reducing the overall social cost—which includes factors like travel time and congestion—by up to 40% compared to approaches that only optimize placement or pricing separately. While motivated by EV charging, the framework is designed to generalize to other domains requiring coordinated resource allocation, such as emergency response and intelligent transportation systems.