A Multiobjective Reinforcement Learning Framework for Equitable Toll Design for Express Lanes

Abstract

Express lanes are commonly used to migrate traffic congestion by providing reliable travel time in exchange for tolls. However, there is a lack of guidance on designing equitable discounts for low-income travelers, which creates barriers to equitable transportation systems. The PI's past study on "Equitable Dynamic Pricing for Express Lanes" addressed some of these gaps by providing guidance for differential tolls and analyzing unintended traffic patterns. However, the framework was limited to single-objective optimization of equity. Furthermore, limitations on open-source algorithms for equity optimization hinder accessibility for researchers and practitioners.

The goal of this implementation-focused research is to develop a multi-objective reinforcement-learning-based optimization of express lane discounts and create an open-source tool for making previous research findings more accessible. In this research, the team will:

(a) design an open-source platform that integrates advances in multi-objective reinforcement learning literature for designing discounts for express lanes,
(b) test the transferability and usefulness of the designed tools across multiple datasets and development platforms, and
(c) conduct a technology transfer to make the tool more accessible for future researchers, practitioners, and policymakers.

The research findings will enable more effective design and optimization of express lane discounts for equitable transportation systems.