Advancing Human-Centered AI for Safe and Resilient Transportation Systems

Research Overview

Our research advances human-centered AI to address transportation risks across the full spectrum, ranging from everyday safety to extreme-event resilience. We focus on understanding, modeling, and mitigating risks that arise from routine transportation operations, including traffic crashes and safety-critical interactions, as well as rare but high-impact disruptions such as hurricanes, flooding, landslides, and the COVID-19 pandemic. 

​​​

Human-Centered AI: What it means in our lab

Human-Centered AI refers to AI systems designed to work with and for people, reflecting human behavior, capabilities, and values. In transportation systems, human road users (e.g., drivers, pedestrians, and cyclists) exhibit complex, dynamic, and sometimes unpredictable behaviors shaped by perception, cognition, decision-making, and environmental context.

 

This behavioral uncertainty represents one of the greatest challenges in improving transportation safety and resilience, as it plays a central role in the formation of transportation risks. It motivates our focus on integrating human factors into AI system design. Human-centered AI therefore recognizes that transportation systems are fundamentally human-driven systems, where understanding and modeling behavioral variability is essential for developing safe and resilient intelligent solutions.

​

To address transportation risks across time scales, our research is organized around two complementary pillars: everyday safety and extreme-event resilience.

 

Pillar 1: Everyday Risk — Transportation Safety

We develop human-centered AI methods to reduce routine crash risk and protect vulnerable road users. Our research emphasizes understanding, modeling, and mitigating risks that arise from dynamic interactions among human road users and AI-enabled systems.

 

Representative research directions include:

• Modeling pedestrian–vehicle interactions in safety-critical situations: Developing deep reinforcement learning frameworks to model dynamic interactions and analyze evasive behaviors between pedestrians and vehicles, particularly in safety-critical conditions (example).

• Generative behavior modeling using large language models (LLMs): Leveraging LLM-based approaches to generate realistic and context-aware human behaviors for transportation safety analysis (example)

• Crash risk prediction incorporating human behavioral factors: Improving crash risk estimation by integrating driver behavior indicators such as hard braking, speeding, and aggressive maneuvers into predictive models (example)

• Human–AI coordination in mixed traffic environments: Designing methods that improve safe negotiation and mutual understanding between human road users and AI-driven systems. (example)

​

The safety research pillar has been supported by funding from the National Science Foundation (NSF), National Academies of Sciences (NAS), Federal Highway Administration (FHWA), and Virginia Department of Transportation (VDOT).

​

Pillar 2: Extreme Risk — Transportation Resilience

We develop human-centered AI methods to support mobility under extreme disruptions, with a strong focus on coastal flooding risks relevant to Hampton Roads and other vulnerable regions. Our research emphasizes understanding, modeling, and mitigating risks that threaten transportation system functionality, accessibility, and recovery during rare but high-impact events.

 

Representative directions include:

• Flood disruption and mobility-impact modeling: Quantifying how flooding affects accessibility, travel reliability, and network performance using, agent-based traffic simulation, data-driven methods, and AI-enabled approaches (example)

• Network vulnerability analysis: Identifying system weaknesses and prioritizing resilience interventions under resource and budget constraints (example)

• Adaptive transportation operations under disruptions: Developing decision-support methods for routing, staging, and operational strategies as conditions evolve during extreme events. (example)

• Recovery support and resilience planning: Designing data-driven approaches to support post-disruption recovery, restoration prioritization, and long-term infrastructure adaptation (example)

​

The resilience research pillar is supported by projects funded by the National Oceanic and Atmospheric Administration (NOAA), with additional institutional support from Old Dominion University (ODU), the Institute for Coastal Adaptation and Resilience (ICAR), and the Civil & Environmental Engineering Visiting Council (CEEVC).

​