Final Report - 510 (Final Report)

Download File

Final_Report_-_510.pdf

Description

Final Report for Project 510 - This project, conducted under the Safety21 University Transportation Center, advanced the safety of heavy-duty vehicles (HDVs) through physics-informed modeling, statistical analysis, sensor deployment, and socio-technical analysis. The research focused on two objectives: developing a Physics-Informed Reinforcement Learning (PIRL) framework for near-miss detection in tractor-trailers, and designing customized safety sensor adoption strategies for small motor carriers. Methods included simulation of tractor-trailer dynamics to estimate safety probabilities in rear, side, and lane-keeping scenarios; probabilistic analysis of crash and inspection datasets to identify high-risk vehicle age–region groups; and survey-based studies of inspection practices modeled with an Inverse Contextual Bandit (ICB) framework. Industry interviews highlighted workforce shortages, data integration challenges, and trust in automation as barriers to adoption. Results showed PIRL outperforms standard Deep Q-Networks (DQN) in identifying near-miss risks, sensor adoption should be prioritized for specific high-risk fleets, and inspector decision-making adapts but remains inconsistent. Findings support tailored sensor programs, workforce training, and data interoperability standards to promote proactive, cost-effective HDV safety.

Citation

Publication Date

None