Deep Reinforcement Learning for Multi-Asset Infrastructure Management Incorporating Traffic Operations Adaptations and Control – Center for Integrated Asset Management for Multimodal Transportation Infrastructure Systems

Final Report (PDF)

Tech Brief (PDF)

Project Title	Deep reinforcement learning for multi-asset infrastructure management incorporating traffic operations adaptations and control
University	Pennsylvania State University George Mason University
Principal Investigator(s)	PI: Kostas Papakonstantinou, Penn State Co-PI: Shelley Stoffels, Penn State Co-PI: Elise Miller Hooks, George Mason University
Funding Source(s) and Amounts Provided (by each agency or organization)	Penn State Federal Share: $126,044 Penn State Match Share: $126,044 George Mason Federal Share: $66,666 George Mason Match Share: $66,666
Total Project Cost	$385,420
Start and End Dates	05/01/2021 - 08/31/2023
Brief Description of Research Project	Concisely, the following objectives are set forth for this project: (i) account for traffic effects in system-level modeling and I&M planning of large infrastructure networks with multiple assets (e.g. pavements, bridges superstructures, substructures, etc.), also incorporating deterioration and inference models under data uncertainty and partial observability; (ii) connect statistical learning and inference approaches to DRL artificial intelligence computational tools to resolve scalability issues in infrastructure management of large networks and portfolios; (iii) assess options for combined I&M actions and traffic control as a means to minimize safety risks and costs, maximize overall network serviceability and availability, and minimize the negative effects of I&M on traffic flows; (iv) validate developed approaches based on existing networks and in-use practices. Overall, this project will uniquely integrate I&M planning and traffic control operations, significantly enhancing currently used tools for infrastructure asset management, offering considerably expanded I&M action choices and possibilities for increased safety and considerable cost savings. The developed framework can provide guidance to DOTs as to how integration of asset management and traffic operations control can be achieved and what the potential benefits can be.