Transportation Planning Based on Reinforcement Learning Policy Intervention

Ruben Glatt | 23-FS-025

Project Overview

In light of the increasing emphasis on electrified transportation spurred by environmental concerns and government initiatives, this study examined the potential for using machine learning, specifically reinforcement learning (RL), in policy decision-making for energy-efficient transportation. A taxonomy to categorize agent-based modeling (ABM) platforms was established and used to evaluate the capabilities of eight such platforms in modern electrified transportation scenarios. BISTRO, a python wrapper to scale simulations of the BEAM transportation modeling and simulation framework, was chosen for further studies and used to train RL algorithms to optimize particular transportation metrics. Using a simplified city scenario as a benchmark, the feasibility of integrating RL with ABM was confirmed, showing a general capability in optimizing desired Key Performance Indicators (KPIs) and providing actionable insights for policy makers. This approach may serve as a foundation for human-in-the-loop decision-making systems and has the potential of enhancing transportation and energy infrastructure planning.

Mission Impact

The project supports the Laboratory's climate and energy security mission, and aimed at building new competencies in energy-efficient mobility and addresses R&D challenges in the HPC Applied to Energy Innovation thrust. It also supports DOE missions in energy, transportation, and environmental security. 

Based on discussions around the project the PI was invited to contribute a chapter on the impact of machine learning on transportation emissions in the Innovation for Cool Earth Forum (ICEF) innovation roadmap 2023.

Publications, Presentations, and Patents

Harris, Donte, Felipe Leno Da Silva, Wencong Su, and Ruben Glatt. " A review on simulation platforms for agent-based modeling in electrified transportation." IEEE Transactions on Intelligent Transportation Systems (2023): accepted. DOI: 10.1109/TITS.2023.3318928.