This research aims to optimize driver decision making in integrated corridors by diverting key cars to Park-And-Ride structures using a Deep Reinforcement Learning agent. I supported this project by handling the DRL training as well as optimizing the simulator (SUMO/OMNETPP) deployment on our servers.

Abstract

The upcoming Connected Vehicles (CV) technology shows great promise in effectively managing traffic congestion and enhancing mobility for users along transportation corridors. Data analysis powered by sensors in CVs allows us to implement optimized traffic management strategies optimizing the efficiency of transportation infrastructure resources. In this study, we introduce a novel Integrated Corridor Management (ICM) methodology, which integrates underutilized Park-AndRide (PAR) facilities into the global optimization strategy. To achieve this, we use vehicle-to-infrastructure (V2I) communication protocols, namely basic safety messages (BSM) and traveler information messages (TIM) to help gather downstream traffic information and share park and ride advisories with upstream traffic, respectively. Next, we develop a model that assesses potential delays experienced by vehicles in the corridor. Based on this model, we employ a novel centralized deep reinforcement learning (DRL) solution to control the timing and content of these messages. The ultimate goal is to maximize throughput, minimize carbon emissions, and reduce travel time effectively.