About

Susu Xu is an assistant professor in the Department of Civil and Systems Engineering at Johns Hopkins University. Her research focuses on mobile sensing, machine learning, urban computing, smart infrastructure systems, and rapid disaster response. Her current projects include developing learning algorithms and incentive mechanisms to improve the efficiency of urban crowdsensing networks and collaboration, creating physics-informed machine learning algorithms to enable smart and fairness-aware urban infrastructure systems, and utilizing multi-sourced, multi-modal sensing and learning to enhance rapid disaster response systems. Her applications target near-real-time disaster information systems for natural hazards such as earthquakes, hurricanes, and wildfires, as well as spatio-temporal urban sensing and data mining related to air pollution, traffic, and noise, along with large-scale infrastructure monitoring of buildings, bridges, and railway tracks.

Research topics

• Computer Science
• Engineering
• Real-time computing
• Computer Security
• Artificial Intelligence
• Transport engineering
• Construction engineering
• Systems engineering
• Forensic engineering
• Environmental science
• Geology
• Knowledge management
• Seismology
• Embedded system

Selected publications

• Learning Dynamic Belief Graphs for Theory-of-mind Reasoning

  arXiv (Cornell University) · 2026-03-20

  preprintOpen accessSenior author

  Theory of Mind (ToM) reasoning with Large Language Models (LLMs) requires inferring how people's implicit, evolving beliefs shape what they seek and how they act under uncertainty -- especially in high-stakes settings such as disaster response, emergency medicine, and human-in-the-loop autonomy. Prior approaches either prompt LLMs directly or use latent-state models that treat beliefs as static and independent, often producing incoherent mental models over time and weak reasoning in dynamic contexts. We introduce a structured cognitive trajectory model for LLM-based ToM that represents mental state as a dynamic belief graph, jointly inferring latent beliefs, learning their time-varying dependencies, and linking belief evolution to information seeking and decisions. Our model contributes (i) a novel projection from textualized probabilistic statements to consistent probabilistic graphical model updates, (ii) an energy-based factor graph representation of belief interdependencies, and (iii) an ELBO-based objective that captures belief accumulation and delayed decisions. Across multiple real-world disaster evacuation datasets, our model significantly improves action prediction and recovers interpretable belief trajectories consistent with human reasoning, providing a principled module for augmenting LLMs with ToM in high-uncertainty environment. https://anonymous.4open.science/r/ICML_submission-6373/

  PublisherDOI

• Wildfire Evacuation Analysis Using Facebook Data: Evidence from Palisades and Eaton Fires

  arXiv (Cornell University) · 2026-01-03

  preprintOpen access

  The growing frequency and intensity of wildfires pose serious threats to communities in wildland-urban interface regions. Understanding evacuation behavior is critical for effective emergency planning. This study analyzes evacuation during the 2025 Palisades and Eaton Fires using high-resolution Facebook data. We propose a comprehensive framework to derive wildfire evacuation-related metrics, including compliance rate, departure timing, delay, origin-destination flows, travel distance, and destination types. A new metric, Damage-Evacuation Disparity Index (DEDI), identifies areas with severe structural damage but low evacuation compliance. Results reveal spatiotemporal heterogeneity: residents closer to the fire evacuated earlier, whereas late or nighttime orders led to lower compliance and longer delays. Contrasting patterns between East and West Altadena further illustrate this disparity. DEDI-identified communities exhibited higher social vulnerability and fire risk. Most evacuations concluded in residential areas, while longer trips concentrated in hotels and public facilities. These findings showcase the Facebook data's potential for data-driven wildfire evacuation planning.

  PublisherDOI

• Wildfire Evacuation Analysis Using Facebook Data: Evidence from Palisades and Eaton Fires

  ArXiv.org · 2026-01-03

  articleOpen access

  The growing frequency and intensity of wildfires pose serious threats to communities in wildland-urban interface regions. Understanding evacuation behavior is critical for effective emergency planning. This study analyzes evacuation during the 2025 Palisades and Eaton Fires using high-resolution Facebook data. We propose a comprehensive framework to derive wildfire evacuation-related metrics, including compliance rate, departure timing, delay, origin-destination flows, travel distance, and destination types. A new metric, Damage-Evacuation Disparity Index (DEDI), identifies areas with severe structural damage but low evacuation compliance. Results reveal spatiotemporal heterogeneity: residents closer to the fire evacuated earlier, whereas late or nighttime orders led to lower compliance and longer delays. Contrasting patterns between East and West Altadena further illustrate this disparity. DEDI-identified communities exhibited higher social vulnerability and fire risk. Most evacuations concluded in residential areas, while longer trips concentrated in hotels and public facilities. These findings showcase the Facebook data's potential for data-driven wildfire evacuation planning.

  PublisherOA PDF

• Bridging bench to bedside: mechanistic elucidation and clinical translation of vasopressin and its analogues for hemorrhagic shock resuscitation

  Shock · 2026-04-03

  articleSenior author

  Hemorrhagic shock remains a leading cause of preventable mortality in civilian and military trauma, despite advances in trauma systems and resuscitation strategies. Vasopressor therapy for early resuscitation is controversial, as conventional catecholamine agents are limited by adverse effects on organ perfusion and uncertain net benefit during active bleeding. Vasopressin and its analogues (notably terlipressin) have emerged as potential alternatives, with unique non-adrenergic pharmacology that enables rapid hemodynamic restoration, reduced transfusion requirements, and organ protection-properties supported by recent molecular and translational research. However, clinical adoption is hindered by critical knowledge gaps, inconsistent trial outcomes, unresolved questions around patient selection and optimal dosing, practical barriers including regulatory approval and cost, and persistent safety concerns such as potential splanchnic hypoperfusion. This review comprehensively synthesizes the pharmacological profiles, cellular and molecular mechanisms, and preclinical/clinical evidence for vasopressin and its analogues in hemorrhagic shock resuscitation. It critically addresses current controversies, methodological limitations of existing research, and key uncertainties in clinical application-including the risks of intestinal ischemia and the challenges of formulary access for terlipressin in the US. We also highlight the ongoing CAVALIER trial (NCT05958342), a pivotal multicenter randomized study investigating early vasopressin use, and outline specific, actionable directions for mechanistic research and precision clinical trials to address unmet needs. By consolidating multidisciplinary evidence in an unbiased framework, this review clarifies the evolving role of vasopressin-based therapies and aims to inform evidence-based, individualized resuscitation strategies to improve outcomes in this high-risk population.

  PublisherDOI

• From Perceptions to Decisions: Wildfire Evacuation Decision Prediction with Behavioral Theory-informed LLMs

  2025-01-01 · 1 citations

  articleOpen accessSenior author

  Evacuation decision prediction is critical for efficient and effective wildfire response by helping emergency management anticipate traffic congestion and bottlenecks, allocate resources, and minimize negative impacts.Traditional statistical methods for evacuation decision prediction fail to capture the complex and diverse behavioral logic of different individuals.In this work, for the first time, we introduce FLARE, short for facilitating LLM for advanced reasoning on wildfire evacuation decision prediction, a Large Language Model (LLM)-based framework that integrates behavioral theories and models to streamline the Chain-of-Thought (CoT) reasoning and subsequently integrate with memory-based Reinforcement Learning (RL) module to provide accurate evacuation decision prediction and understanding.Our proposed method addresses the limitations of using existing LLMs for evacuation behavioral predictions, such as limited survey data, mismatching with behavioral theory, conflicting individual preferences, implicit and complex mental states, and intractable mental statebehavior mapping.Experiments on three postwildfire survey datasets show an average of 20.47% performance improvement over traditional theory-informed behavioral models, with strong cross-event generalizability.

  PublisherOA PDFDOI

• From Image Generation to Infrastructure Design: a Multi-agent Pipeline for Street Design Generation

  ArXiv.org · 2025-09-05

  preprintOpen accessSenior author

  Realistic visual renderings of street-design scenarios are essential for public engagement in active transportation planning. Traditional approaches are labor-intensive, hindering collective deliberation and collaborative decision-making. While AI-assisted generative design shows transformative potential by enabling rapid creation of design scenarios, existing generative approaches typically require large amounts of domain-specific training data and struggle to enable precise spatial variations of design/configuration in complex street-view scenes. We introduce a multi-agent system that edits and redesigns bicycle facilities directly on real-world street-view imagery. The framework integrates lane localization, prompt optimization, design generation, and automated evaluation to synthesize realistic, contextually appropriate designs. Experiments across diverse urban scenarios demonstrate that the system can adapt to varying road geometries and environmental conditions, consistently yielding visually coherent and instruction-compliant results. This work establishes a foundation for applying multi-agent pipelines to transportation infrastructure planning and facility design.

  PublisherOA PDFDOI

• When Recovery Becomes Infeasible: A Markov Model of Housing Abandonment Risk in Flood-Prone Areas

  2025-12-19

  preprintOpen access

  Floods can undermine long-term community viability by depressing housing markets and triggering property abandonment cycles. This study estimates the risk of housing abandonment by integrating traditional flood-risk frameworks with the concept of sub-replacement–a condition where the cost of repairing a house exceeds its market value. We propose a new risk metric that identifies whether a house enters a sub-replacement condition within a given time horizon. Our stochastic model is a time–homogeneous, discrete-time Markov process that incorporates flood hazard, housing exposure, physical vulnerability, and housing market dynamics. We apply the model to two U.S. communities–Pascagoula, Mississippi, and McGregor, Florida. These two communities exhibit similar flood hazard, exposure, and building vulnerability, but markedly different housing market conditions. Despite comparable Average Annual Losses (AAL), the number of houses expected to experience sub-replacement within the next three decades is twenty five times larger in Pascagoula than in McGregor. We also find that the FEMA 50\% rule–which mandates elevation when repair costs exceed 50\% of a home’s market value–reduces AAL by approximately 70\%, but increases sub-replacement risk in areas with depressed housing markets. This risk is especially concerning in Pascagoula, where lower housing prices increase the number of houses expected to enter sub-replacement in the next three decades by a factor of eight. Our findings show that incorporating housing market conditions into flood risk analysis is important for anticipating long-term recovery trajectories and prevent downward spirals of disinvestment and population loss.

  PublisherOA PDFDOI

• Spatially-Heterogeneous Causal Bayesian Networks for Seismic Multi-Hazard Estimation: A Variational Approach with Gaussian Processes and Normalizing Flows

  ArXiv.org · 2025-04-05

  preprintOpen accessSenior author

  Post-earthquake hazard and impact estimation are critical for effective disaster response, yet current approaches face significant limitations. Traditional models employ fixed parameters regardless of geographical context, misrepresenting how seismic effects vary across diverse landscapes, while remote sensing technologies struggle to distinguish between co-located hazards. We address these challenges with a spatially-aware causal Bayesian network that decouples co-located hazards by modeling their causal relationships with location-specific parameters. Our framework integrates sensing observations, latent variables, and spatial heterogeneity through a novel combination of Gaussian Processes with normalizing flows, enabling us to capture how same earthquake produces different effects across varied geological and topographical features. Evaluations across three earthquakes demonstrate Spatial-VCBN achieves Area Under the Curve (AUC) improvements of up to 35.2% over existing methods. These results highlight the critical importance of modeling spatial heterogeneity in causal mechanisms for accurate disaster assessment, with direct implications for improving emergency response resource allocation.

  PublisherOA PDFDOI

• Permanent Relocation Into and Out of Areas Exposed to Natural Hazards: a Multidisciplinary Review of the Literature

  2025-03-27 · 2 citations

  reviewOpen access

  This article examines the long-term impacts of natural hazards caused by patterns of relocation into and out of hazard-exposed communities. We address two main questions: (1) what factors influence permanent relocation decisions in hazard-exposed communities? (2) What are the effects of relocation on the socio-economic and demographic characteristics of these communities? To answer these questions, we review studies on theoretical frameworks, empirical analyses, and simulation-based models. Relocation outcomes result from a complex interplay of household characteristics (e.g., wealth, risk perception, place attachment), community characteristics (e.g., economic opportunities, essential services), and government interventions (e.g., collective risk-reduction measures). The reviewed studies report mixed findings on demographic and socio-economic changes associated with permanent relocation. Large-scale analyses suggest that natural hazards have limited effects on pre-existing population trends, while more granular studies show that specific hazards—such as coastal flooding and sea level rise—can alter local dynamics. Effects on communities socio-economic characteristics also vary. Some communities experience post-hazard gentrification, while others face deepened vulnerabilities, with declining property values trapping residents in high-risk areas. We further review simulation-based models that examine hazard-related relocation and the socio-economic changes it can produce. These models often focus on specific aspects, such as individual decision-making, housing markets, or recovery patterns, without integrating all relevant factors. Finally, we identify key research gaps, including the need for more long-term studies on socio-economic changes in hazard-exposed communities, and greater focus on chronic, low-intensity hazards like tidal flooding.

  PublisherDOI

• From Perceptions to Decisions: Wildfire Evacuation Decision Prediction with Behavioral Theory-informed LLMs

  ArXiv.org · 2025-02-24 · 1 citations

  preprintOpen accessSenior author

  Evacuation decision prediction is critical for efficient and effective wildfire response by helping emergency management anticipate traffic congestion and bottlenecks, allocate resources, and minimize negative impacts. Traditional statistical methods for evacuation decision prediction fail to capture the complex and diverse behavioral logic of different individuals. In this work, for the first time, we introduce FLARE, short for facilitating LLM for advanced reasoning on wildfire evacuation decision prediction, a Large Language Model (LLM)-based framework that integrates behavioral theories and models to streamline the Chain-of-Thought (CoT) reasoning and subsequently integrate with memory-based Reinforcement Learning (RL) module to provide accurate evacuation decision prediction and understanding. Our proposed method addresses the limitations of using existing LLMs for evacuation behavioral predictions, such as limited survey data, mismatching with behavioral theory, conflicting individual preferences, implicit and complex mental states, and intractable mental state-behavior mapping. Experiments on three post-wildfire survey datasets show an average of 20.47% performance improvement over traditional theory-informed behavioral models, with strong cross-event generalizability. Our complete code is publicly available at https://github.com/SusuXu-s-Lab/FLARE

  PublisherOA PDFDOI

Recent grants

• Rapid, Scalable, and Joint Assessment of Seismic Multi-Hazards and Impacts: From Satellite Images to Causality-Informed Deep Bayesian Networks

  NSF · $392k · 2024–2025

Frequent coauthors

• Hae Young Noh

  30 shared

• Xinlei Chen

  Peng Cheng Laboratory

  28 shared

• Xuechun Li

  Johns Hopkins University

  14 shared

• Carlee Joe‐Wong

  9 shared

• Shijia Pan

  University of California, Merced

  8 shared

• David J. Wald

  United States Geological Survey

  8 shared

• Chenguang Wang

  Stony Brook University

  7 shared

• Xiao–Ping Zhang

  University Town of Shenzhen

  6 shared

Awards & honors

• NSF CAREER Award
• American Society of Mechanical Engineers (ASME) Structural H…
• International Conference on Machine Learning and Application…
• Champion of NeurIPS Adversarial Vision Challenge
• MIT’s Civil and Environmental Engineering Rising Star award