Poster Sessions

ABSTRACT

Understanding crash patterns and identifying effective countermeasures have always been challenging as it requires a deeper look into the roadway, behavioral, and environmental contexts that shape crash outcomes. This study proposes a novel data-driven methodology that combines machine learning, clustering, and spatial risk scoring to evaluate and prioritize countermeasures at both intersection and corridor levels. The approach was designed to be practical and transferable so that cities and counties can easily apply it under the Safe System Approach framework.

This methodology was implemented in USDOT-funded Safe Streets for All (SS4A) projects for Eureka and El Dorado, Kansas, successfully identifying locally relevant countermeasures and revealing behavioral and roadway risk patterns that traditional screening methods often overlook. Crash data from local transportation networks were analyzed to identify variables most associated with crash severity using Random Forest (RF). The selected factors—covering roadway type, control type, lighting, surface condition, and driver behaviors were used to group similar crashes through several unsupervised algorithms. Among those, K-Modes provided the most reliable representation of categorical crash attributes. Each resulting cluster represents a unique crash context or “crash typology,” where crashes within a cluster share similar characteristics while remaining distinct from other clusters. Detailed examination of cluster profiles allowed identifying the dominant contributing factors, crash scenarios, and overrepresented conditions for each cluster. Countermeasures were then developed for each cluster by combining insights from relevant literature, FHWA and NCHRP guidance, and practical engineering judgment and field experience.

Moreover, a severity-based scoring system was developed to quantify crash risk by assigning greater importance to more severe crashes while still considering less severe and property-damage-only crashes. The weighting structure was derived from crash cost relationships reported in the Highway Safety Manual and crash severity cost estimates used in Kansas Department of Transportation safety analyses, and was further calibrated with local crash characteristics. Fatal and disabling injuries were combined because both often share similar contributing conditions. To capture the spatial dimension of risk, DBSCAN clustering was used to aggregate crashes at intersections and roadway segments separately. The cumulative risk score at each location was normalized by its crash proportion within the same facility type to create a ranked High-Injury Network (HIN) highlighting the top high-risk intersections and corridors.

Integrating the cluster-level profiles with the HIN helped determine which crash types dominate each high-risk location. Moreover, this method allows to not only evaluate the countermeasures for HIN but also prioritize the countermeasures for each intersection and segment. For instance, if a site showed 50%, 30%, and 20% of crashes from Clusters 1, 2, and 3 respectively, countermeasure priorities would follow the same order—giving highest priority to strategies effective for Cluster 1, followed by Clusters 2 and 3. This proportional approach ensures that safety improvements directly reflect the dominant crash patterns present at each site, leading to more targeted and efficient countermeasure implementation. The approach provides a scalable, transferable, and reproducible framework for cities and counties seeking to implement data-informed, human-centered safety improvements consistent with Vision Zero and the Safe System principles.

PRESENTER

Saumik Sakib Bin Masud

Dr. Saumik Sakib Bin Masud is a Traffic Engineer at JEO Consulting Group specializing in transportation safety, the Safe System Approach, and data-driven decision making. His work focuses on developing Safe Streets and Roads for All (SS4A) Safety Action Plans, conducting highway safety and operational analyses, evaluating roadway and interchange alternatives, and applying predictive safety methods, machine learning, and advanced analytics to support evidence-based transportation planning, infrastructure investment, and policy decisions. He earned his Ph.D. in Transportation Engineering from the University of Kansas, where his research advanced the understanding of human factors and automated driving systems through machine learning and driving simulation. He continues to conduct applied research in transportation safety, intelligent transportation systems, and artificial intelligence, bridging research and practice to improve roadway safety and mobility. Dr. Masud is an active member of the ITE Safety Council, the ITE Vision Zero Standing Committee, and ASCE, and regularly presents his work at national and international conferences while contributing to innovative, data-driven transportation safety solutions.

Amirmohammad Sadeghnejad, Li Zhao, University of Nebraska-Lincoln 

More Information Coming Soon!

ABSTRACT

Traditional seat belt observational surveys provide statistically representative estimates of restraint use but require substantial staffing, travel, and data collection resources. This study develops and evaluates a calibration framework for estimating population-representative seat belt use with routinely collected crash report data. Arkansas was used as a case study to develop regression-based calibration models using nine years (2013–2019 and 2022–2023) of crash records and observational seat belt survey data collected following National Highway Traffic Safety Administration (NHTSA) protocols. Data were aggregated into 111 county-year observations. Ordinary least squares (OLS) calibration models were estimated with observed seat belt use as the dependent variable and crash-derived restraint rates, vehicle composition, demographic characteristics, injury severity, and urbanicity as candidate predictors. Temporal stability was evaluated using rolling-window validation.

The baseline model using only crash-based belt use as an independent variable to predict population representative belt use achieved an adjusted R² of 34.65%, Root Mean Square Error (RMSE) of 0.0599, and Mean Absolute Percent Error (MAPE) of 6.29%. The final model retained no-injury crash-based seat belt use, sedan share, young share, and urbanicity, improving performance to an adjusted R² of 52.37%, an RMSE of 0.0438, and a MAPE of 4.53%. Rolling-window validation indicated that prediction error increased by approximately 0.41 percentage points in MAPE for each additional year beyond the calibration period. This study presents a replicable calibration framework for estimating population-representative seat belt use with crash-derived restraint data and county-level contextual variables. To the authors’ knowledge, no prior study has developed a regression-based calibration framework linking crash-reported and observed seat belt use rates. The methodology provides transportation agencies with a supplemental approach for estimating seat belt use using existing crash reporting systems and may support continuous monitoring between observational survey years.

PRESENTER

Karla Diaz-Corro

Karla Diaz-Corro is a traffic engineer and Ph.D. student in Civil Engineering at the University of Arkansas, specializing in transportation safety, technology, and system performance. She earned her bachelor’s and master’s degrees in Civil Engineering from the University of Arkansas in 2018 and 2020, respectively. Her research focuses on developing innovative, data-driven solutions to improve the safety and efficiency of transportation systems. Karla leads research projects for the Arkansas State Police Highway Safety Office, including the Traffic Safety Campaign Awareness Survey and the statewide Seat Belt Use Study. An active member of the Institute of Transportation Engineers, she serves as Co-Chair of both the MOVITE Student Activities Committee and the Women in ITE Committee. Through her research, leadership, and mentoring, Karla is committed to advancing transportation engineering while supporting women, students, and emerging professionals in the field.

Samuel Durairaj, University of Kansas Medical Center 

More Information Coming Soon!

ABSTRACT

Teen crash involvement remains a major traffic safety concern despite the widespread implementation of Graduated Driver Licensing (GDL) systems. Family support is typically viewed as protective because it can increase access to driver education, supervised practice, vehicles, and insurance. However, this study examines whether such support may also create an unintended safety tradeoff by accelerating adolescents’ progression toward independent driving. Using survey data from 469 licensed young adults in the United States, we tested whether family support and family barriers were associated with crash involvement through two licensing pathways: permit delay, representing developmental timing, and GDL duration, representing time spent in supervised or restricted driving stages. Bootstrap mediation models were used to estimate indirect effects, and Cox proportional hazards models were used to examine time to first crash after licensure.

The results suggest that licensure timing was the dominant pathway linking family factors to crash involvement. Family support factors, including parent-paid insurance, supervised practice, and vehicle access, were associated with earlier permit acquisition, which in turn was linked to greater crash involvement. In contrast, family barriers were associated with delayed licensing progression and lower crash involvement. The permit-delay pathway was stronger than the GDL-duration pathway, indicating that developmental timing may play a larger role in early driving safety than supervised practice duration alone. These findings reveal a family support paradox: well-intended family resources may improve access to licensure while unintentionally increasing crash exposure by enabling earlier independent driving.

PRESENTER

Yuxi Shen

Yuxi Shen is a Ph.D. student in Civil and Environmental Engineering at Michigan State University. Her research focuses on transportation safety, human factors, and data-driven approaches to improving roadway safety and mobility. Her work examines topics including young driver behavior, roadway infrastructure, emergency medical services, and interactions between drivers and emerging vehicle technologies. Her research has been published in leading transportation journals, including Transportation Research Part D and the Journal of Transport Geography. Through her research, Yuxi aims to better understand the factors that influence transportation safety and to develop practical, evidence-based solutions that support safer roadway systems and informed transportation decision-making.

ABSTRACT

Work zones are one of the most challenging traffic disruptions for agency operators to manage, given their varied durations, impacts, and externalities. Road agencies with large geographic footprints have many projects with work zones—large or small—across their system but struggle to maintain a real-time understanding of their impacts. Work zones not designed or managed to reflect real-time traffic can trigger unnecessary traffic queues, causing brooding frustrations among motorists. Longer-term work zones may see traffic divert onto alternate routes, regardless of if queues or delays exist within the work zone, sometimes causing undesirable queues on those alternate routes due to the new travel demand exceeding the available road supply. Traditional methods for monitoring this required extensive investment in traffic data devices, temporary surveillance assets, and routine field personnel to visit each site to monitor work zone performance, a proposition too costly for most organizations.

Fortunately, the rise of mobile communications and internet-of-things devices has created opportunities to monitor work zone impacts without the burdensome costs. Big data software tools have emerged to remotely collect, process, and report systemwide traffic conditions. Data scientists have taken this a step further, developing tools to quickly find meaning within these datasets. These tools allow agencies to identify speeding hotspots, estimate changes in turning movements at signalized intersections, and examine traffic signals that see increases in undesirable split failures. Additionally, new tools that remotely access dash camera video feeds and work zone smart device locations leverage artificial intelligence to help “tell the story” of what is happening at a given work zone. All these tools come together to give agency decisionmakers easy access to insights, allowing them to quickly make informed decisions to respond to undesired conditions.

This presentation highlights several case studies where big data and supporting industry tools have been used to identify problems and improve decision-making. A few examples:

• Utilizing dashboard cameras to assess the real-time state of the work zone (lane placements, widths, traffic performance), as well as to provide visibility to areas of the road network that lacks ITS cameras
• Leveraging historical heat maps to quickly evaluate citizen complaints of traffic signal delays within work zones that see isolated queue spikes at key times during the week.
• Responding to traffic system breakdowns by isolating the cause of the new traffic trend, often a work zone stage change that happens on a different facility.
• Integrating connected arrow board and smart work zone device data to understand staging changes, lane closures, and work crew movements — providing context that explains unexpected spikes or improvements in traffic performance.

PRESENTERS

Christos Achillides

Christos Achillides is a Vice President with over 21 years of Traffic Engineering experience. Christos' experience includes signal system optimization studies using Signal Performance Measures (SPM), ITS/transportation systems design and operations, MOT plans for construction projects, system integration, and signal warrant studies.

Adam Danczyk

Adam is a principal engineer with over 20 years of ITS planning, design, and deployment experience. He has been involved in a wide variety of technical matters involving high-profile transportation technology topics across the United States, including truck parking systems, connected and automated vehicles, and leveraging big data platforms to deliver insights to influence business decisions in traffic operations.

ABSTRACT

As Level 4 automated vehicles become more common in mixed traffic, understanding how human drivers respond to them is increasingly important. This study examines whether human-driven vehicles exhibit different car-following behaviors when following Level 4 automated vehicles compared with human-driven leaders. The analysis used 1,140 arterial-road car-following segments, including 891 human-following-human and 249 human-following-Level 4 interactions. A pooled Gaussian Hidden Markov Model was first applied to high-resolution trajectory data to identify latent longitudinal-interaction states. Conventional trajectory-based measures and HMM-derived temporal features were then combined to develop an interpretable behavioral taxonomy using principal component analysis and Gaussian mixture modeling.

The final taxonomy included four patterns: highly reactive closing following, low-gap stop-and-go following, routine adaptive following, and stable long-gap following. The raw distribution of these patterns differed significantly by leader type (χ² = 41.86, p < 0.001; Cramér’s V = 0.192). Repeated balanced sampling confirmed that this association was not driven by unequal group sizes; all 1,000 iterations remained statistically significant, with a median Cramér’s V of 0.223. After controlling for baseline follower speed, initial gap, and initial relative speed, leader type remained jointly associated with taxonomy membership (p = 0.031). However, the association was attenuated after broader leader-motion characteristics were included.

The findings show that human responses to Level 4 automated vehicles are heterogeneous and partly shaped by the motion context created by the leader. The proposed taxonomy provides a useful framework for mixed-traffic simulation and automated-vehicle evaluation.

PRESENTER

Bahareh Bakhti

My name is Bahareh Bakhti, and I am a fourth-year Ph.D. student in Civil Engineering at the University of Kansas. My primary research area is transportation safety, with a particular interest in applying data-driven and computational methods to understand and improve roadway safety. Currently, my research focuses on the safety of automated vehicles and their interactions with human-driven vehicles. I am particularly interested in studying driving behavior, car-following interactions, reaction time, and surrogate safety measures using real-world trajectory data and traffic simulation. Through my research, I aim to better understand the safety implications of automated vehicle technologies and contribute to the development of safer and more reliable transportation systems.

ABSTRACT

With the growing adoption of autonomous driving technologies, human-driven vehicles (HVs) are increasingly expected to coexist with autonomous vehicles (AVs) on public roads. Consequently, enabling AVs to interact seamlessly with surrounding human drivers has become a critical challenge, highlighting the need for effective human–robot interaction strategies. A promising approach inspired by psychology is the use of Social Value Orientation (SVO), which enables AVs to quantify cooperation levels and social preferences. While limited AV control frameworks have incorporated human–robot interaction concepts, most assume a constant SVO, resulting in fixed driving behaviors that may not generalize well across diverse and complex traffic scenarios. As a result, existing controllers may fail to appropriately balance individual efficiency with their impacts on surrounding traffic. 

Given the highly dynamic and heterogeneous nature of traffic flow, the optimal behavior of an AV is inherently context dependent. Therefore, the optimal SVO of an AV may vary between altruistic and egoistic behaviors depending on prevailing traffic conditions. To address this gap, this study proposes an optimal autonomous driving framework with dynamic SVO adaptation for AVs. The proposed controller aims to improve overall traffic conditions by enhancing traffic stability and safety while reducing fuel consumption (FC). An optimal control problem is formulated based on the utilities of both the AV and its following HV, incorporating the AV’s FC and the speed fluctuations of the following HV. Leveraging Pontryagin’s Minimum Principle, the controller determines the optimal acceleration inputs and SVO angles that minimize the objective function. The resulting dynamic SVO enables the AV to adapt its level of cooperation according to traffic conditions and interactions with surrounding vehicles. The proposed control strategy is evaluated through a comprehensive set of simulations using synthetic trajectories, real-world highway data, and urban driving scenarios covering diverse traffic environments. 

PRESENTER

Jose Carlos Acedo

Jose Carlos Acedo Aguilar was born and raised in Ciudad Juarez, Chihuahua, Mexico. He started his college studies at the Institute of Technology in Ciudad Juarez (ITCJ) before transferring to the University of Texas at El Paso (UTEP). He is currently a Ph.D. candidate in the Department of Electrical and Computer Engineering at The University of Texas at El Paso. He received his bachelor's degree in mechanical engineering and master's in electrical engineering from the same institution. His research focuses on intelligent transportation systems and control systems. He has received multiple awards, including the Best Paper Runner-Up Award of 2024 IEEE Forum for Innovative Sustainable Transportation Systems.

ABSTRACT

According to the Highway Capacity Manual (HCM), Level of Service (LOS) F occurs when demand exceeds capacity or when the selected service measure exceeds a certain threshold. However, no further clarification is provided regarding the degree of congestion that may exist within this LOS category. This study aims to assist transportation agencies in evaluating the severity of congestion and assigning subcategories within LOS F. A freeway segment along Interstate 5 (I-5) in California was analyzed. Congested traffic conditions were identified using the HCM breakdown criteria, and Principal Component Analysis (PCA) and the K-means algorithm were applied to the dataset to define clusters. The Van Aerde model and Support Vector Machine

(SVM) were then used to determine the thresholds for each cluster. These thresholds were calculated based on density, aligning with the HCM performance measures for uninterrupted flow segments. Cluster robustness was evaluated using the silhouette score, which was 0.59 for both the training and validation datasets. Jaccard overlap was also calculated to assess the consistency between the training and validation datasets and was

very close to 1 across all intervals. Additionally, the lower threshold of the first cluster, representing moderate congestion, was 33 passenger cars per mile per lane (pc/mi/ln), which closely aligns with the updated LOS F threshold reported in NCHRP Report 1038. The results of this study can be used to subclassify LOS F and distinguish among different levels of congestion.

PRESENTER

Mahgam Tabatasbaei

Mahgam Tabatabaei is a Ph.D. student in Civil Engineering at the University of Kansas, specializing in Transportation Engineering. Her research focuses on traffic flow theory, freeway operations, and the application of machine learning to transportation systems. She has presented her research at national and international conferences. In addition to her research, she is actively involved in student leadership and currently serves as Co-President of the KU Institute of Transportation Engineers (ITE) student chapter. She enjoys collaborating with researchers and transportation professionals to advance innovative and practical solutions for transportation systems.

ABSTRACT

Stop-and-go waves degrade the safety, comfort, and energy efficiency of congested traffic. Large language models (LLMs) offer semantic, human-interpretable reasoning about how such waves develop, but their inference latency and lack of formal safety guarantees have kept them out of the longitudinal control loop. This study proposes LLM-λ-IDM, a framework that admits LLM reasoning into car-following control while making safety independent of the LLM itself. The LLM controls exactly one quantity: a scalar sensitivity coefficient λ that scales the interaction term of the Intelligent Driver Model (IDM), acting as an interpretable gap-keeping aggressiveness dial. Collision immunity holds structurally for every positive value of λ, and an execution loop running at the native ≈30 Hz rate of the field data projects each request onto a provably safe domain built from control-barrier-function, time-to-collision, and comfort conditions, so that a late, missing, or malformed reply cannot compromise safety. A 2 Hz semantic loop renders the traffic state in natural language for the LLM. The closed-loop system is evaluated on a six-vehicle platoon replaying 551 s of 30 Hz field trajectories from the most string-unstable recording of the Arizona Ring Experiments Dataset, under an eight-dimension evaluation protocol. Controlling a single vehicle in the third platoon position reduces head-to-tail disturbance amplification from 1.25 to 1.15, cuts time-exposed time-to-collision by 53% and total stop time by 31%, and lowers root-mean-square jerk by 21%, while mean follower speed changes by less than 0.05% and the 5th-percentile gap widens by 21%. The results demonstrate a transferable design pattern for certifiable language-model supervision in safety-critical vehicle control.

PRESENTER

Shuxia Pang

Shuxia Pang is a Ph.D. student in the Department of Civil Engineering at Saint Louis University, where she is advised by Dr. Tianyi Li. Her research lies at the intersection of traffic flow theory, automated vehicle control, and artificial intelligence, with a current focus on how large language models can be embedded into microscopic car-following controllers without sacrificing formal safety guarantees. She is also interested in the spatial analysis of urban transit systems, including geographically weighted regression studies of built-environment effects on metro–bus transfer ridership. Her work has appeared in Artificial Intelligence for Transportation, and she has presented at regional and national transportation research venues. Her long-term goal is to develop interpretable, deployable control and planning methods that make congested urban traffic safer, smoother, and more energy-efficient.

ABSTRACT

Connected and autonomous vehicles (CAVs) offer a promising means to improve the efficiency and stability of mixed traffic through both their own longitudinal motion and their indirect influence on upstream human-driven vehicles (HVs). Existing CAV control studies have largely relied on prescribed microscopic car-following models or macroscopic traffic flow models to characterize upstream HV responses. However, these assumptions become

restrictive when traffic is heterogeneous, partially observable, and governed by uncertain human behaviors. To address this gap, we propose a learningbased framework that leverages a controlled CAV as a moving downstream boundary to optimize upstream traffic flow without explicitly prescribing the car-following dynamics of individual HVs. The upstream road segment is discretized into multiple cells, where sparsely distributed connected vehicles

(CVs) serve as mobile probes providing partial observations of local traffic states, such as speed. A cell-level macroscopic dynamics model is trained using microscopic trajectory data generated in SUMO to predict the evolution of traffic density and speed, with flow inferred from the predicted states. The model is conditioned on sparse CV probe observations, current cell-level state estimates, and the downstream CAV control action. Based on the learned dynamics, the controlled CAV optimizes its speed trajectory to regulate upstream traffic states and improve traffic efficiency. Simulation experiments under varying traffic demands and CV penetration rates demonstrate the effectiveness of the proposed upstream-aware CAV control strategy in improving traffic efficiency compared with self-centered CAV control baselines.

PRESENTER

Xinyuan Li

Xinyuan Li is a Ph.D. student in Civil Engineering at the University of Kansas. His research interests include intelligent transportation systems, connected and automated vehicles, traffic flow modeling, and learning-based traffic control.

Yuhui Liu, Saint Louis University 

More Information Coming Soon!

ABSTRACT

We present a priority-aware intelligent lane change advisory system based on multi-agent federated reinforcement learning, namely PALCAS, for autonomous vehicles (AVs). While existing lane-change approaches typically focus on single-agent systems or centralized multi-agent systems, we introduce a federated reinforcement learning-based multi-agent lane change system prioritizing lane changing based on vehicle destination urgency. kut

PALCAS incorporates a novel priority-aware safe lane-change reward function to enable judicious lane-change decisions in both mandatory and discretionary scenarios. PALCAS leverages the parameterized deep Q-network (PDQN) algorithm to facilitate effective cooperation among agents, enabling both lateral and longitudinal motion controls of AVs. Extensive simulations conducted using the SUMO traffic simulator and Mosaic V2X communication framework demonstrate that PALCAS significantly improves traffic efficiency, driving safety, comfort, destination arrival rates, and merging success rates compared to baseline methods.

PRESENTER

Yassine Ibork

Hello, my name is Yassine Ibork. I am a PhD researcher in computer science from Morocco, specializing in artificial intelligence for autonomous driving. My work focuses on developing intelligent systems capable of learning, adapting, and making reliable decisions in complex driving environments. I am particularly interested in reinforcement learning, federated learning, machine learning, and computer vision. Through my doctoral research, I aim to contribute to safer, more efficient, and trustworthy autonomous vehicles. I am motivated by the real-world impact of artificial intelligence, I hope to advance intelligent mobility and help shape the future of responsible autonomous transportation through innovative and meaningful research.

ABSTRACT

High-level autonomous vehicles (AVs) have been trialed in numerous urban areas worldwide, promising enhanced safety and more convenient travel. However, this technology has not been fully realized in rural America. Rural roads present different risks, with 45 percent of fatalities occurring on them and especially high rates of single-vehicle run-off-the-road crashes. For AVs to succeed in rural areas, the needs of rural road users must be understood and met. This research evaluates current and future rural user needs for AVs through a stated preference survey, which collects demographic information, travel behavior, AV experience, and opinions on future adoption. So far, 243 responses have been collected. The data is analyzed to identify demographic predictors of AV acceptance and quantify the strength of those associations. This research aims to inform policymakers, transportation professionals, and AV manufacturers as they prepare rural areas for autonomous vehicles.

PRESENTER

Logan Pittman

Logan Pittman is a civil engineering graduate student at KU. He graduated with his BS in Applied Engineering from Wichita State in 2024 and expects to graduate with his MS this winter. He simultaneously works full time for the Kansas Department of Transportation as a civil engineer-in-training and hopes to attain his professional engineer license soon. Logan's ambition is to help create a transportation network that prioritizes safety, especially for pedestrians and cyclists. In his free time, Logan enjoys weightlifting and playing board games with friends.

ABSTRACT

Heavy-duty electric vehicles (HDEVs) are now technically viable for long-haul freight , but deployment is limited by insufficient megawatt-scale charging infrastructure. Unlike light-duty EVs, HDEVs require strategically placed fast-charging stations enabling drivers to recharge within mandatory rest periods. This challenge is acute for small and medium-sized fleets lacking capital for dedicated infrastructure. Emerging coalition-based models propose independent operators investing in semi-public infrastructure through shipper contracts. Our work is motivated by a pilot deploying HDEV charging along Interstate 10. Designing such networks is a complex bilevel optimization problem: the operator (principal) determines station locations, capacities, and time-of-use pricing while anticipating heterogeneous fleets’ (agents) routing and charging responses. Insufficient capacity or poor pricing induces costly queues; overbuilding is capital-intensive. We develop a bilevel framework integrating infrastructure design, dynamic pricing, and fleet equilibrium within a time-expanded network (TEN), and introduce augmented Lagrangian reformulation and neural network-based value function approximation.

PRESENTER

Mansimran Singh

I am a Ph.D. student in Applied Mathematical and Computational Sciences at the University of Iowa. My research focuses on operations research, transportation, and sustainable logistics, with particular interest in shared charging infrastructure for heavy-duty electric vehicle fleets. I develop optimization and equilibrium models that examine how charging capacity, time-of-use pricing, congestion, and decentralized fleet decisions interact in long-haul freight networks. My broader research interests include network design, stochastic optimization, reverse logistics, and dynamic decision-making. Before beginning my doctoral studies, I taught mathematics at the University of Delhi. I have also received the University of Iowa College of Liberal Arts and Sciences Outstanding Teaching Assistant Award. Through my research, I aim to develop practical analytical tools that support more efficient and sustainable transportation and logistics systems.

Samiha Karim Subah, Richa Bhattarai, Arif Mohaimin Sadri, University of Oklahoma

More Information Coming Soon!

ABSTRACT

Worn lane markings weaken the guidance that drivers and lane-keeping systems rely on and are associated with lane-departure crashes, so degraded lines must be identified and restriped before they become hazards. Vision-language models can read marking condition from dashcam imagery, but their reliability under uneven illumination is not established. This study evaluates GPT-5.4 and Claude Opus 4.8 on US-75 corridor frames where intact lane markings pass from daylight into bridge-underpass shadow, using the shadow as a natural experiment that separates the effect of lighting from that of wear. Three prompts are compared, and each frame is judged three times. Illumination is the main obstacle: across all prompts, intact markings in shadow are flagged as worn several times more often than in good light. The few-shot prompt gives the best balance, and routing only the frames where repeated queries disagree to human review yields a workflow suited to maintenance screening. 

PRESENTER

Tianyang Cui

Tianyang Cui is a second-year PhD student in Mechanical Engineering at the University of Texas at Dallas, supervised by Dr. Zejiang Wang. His primary research focuses on scenario generation for intelligent transportation systems, developing methods to create and test-driving scenarios for safety evaluation. More broadly, his work applies artificial intelligence and computer vision to transportation problems, including the vision-language model approach to lane-marking condition assessment presented in this poster. He is interested in developing practical AI tools that advance transportation safety and infrastructure monitoring.

ABSTRACT

Soil slope failures occasionally occur along transportation infrastructure, particularly after intense or prolonged rainfall. Selecting an appropriate mitigation method requires consideration of both stability improvement and practical constraints, such as construction footprint, material availability, and cost. This study evaluated four mitigation methods for potential soil slope failures: slope flattening, geosynthetic reinforcement, stone replacement, and lightweight aggregate replacement. A two-dimensional numerical model was developed based on an embankment slope failure along K-18 in Russell County, Kansas. The model was validated using an analytical infinite slope solution and the interpreted failure surface from the KDOT case history. A fully saturated slope condition was assumed to represent a conservative condition and provide a consistent basis for comparing the mitigation methods. Factors of safety and failure mechanisms were evaluated using the shear strength reduction technique. Parametric analyses were performed to examine how the layout, material properties, and replacement extent of each mitigation method affected slope stability.

The numerical results showed that all four mitigation methods improved slope stability, although their controlling mechanisms differed. Slope flattening improved stability by reducing the driving shear stress, while geosynthetic reinforcement modified the critical failure surface and improved internal stability. The effectiveness of stone replacement was mainly controlled by the bottom width of the replacement zone. The effectiveness of lightweight aggregate replacement depended not only on the replacement extent but also on whether the replacement zone covered critical regions controlling the development of the failure surface. A cost analysis was conducted based on the method-specific expenses required to achieve selected target factors of safety. Slope flattening had the lowest direct method-specific cost under the adopted assumptions, although its applicability may be limited by right-of-way requirements. Geosynthetic reinforcement became more competitive at a higher target factor of safety, whereas stone and lightweight aggregate replacement showed greater cost increases because of the larger required replacement volumes. These results provide preliminary guidance for selecting mitigation alternatives for slope failures under similar field conditions.

PRESENTER

Haohua Chen

Haohua is a Ph.D. student in civil engineering at the University of Kansas, specializing in geotechnical engineering. His research focuses on transportation earth structures, with an emphasis on clarifying mechanisms, enhancing serviceability, and mitigating failures under different ground, loading, and environmental conditions. He integrates soil mechanics, numerical modeling, and engineering case analysis to develop practical insights and methods for safer and more resilient infrastructure. His research topics include rainfall-induced slope failures, embankments over soft ground, ground improvement, geosynthetic reinforcement, and deformation-related stability assessment. His current work evaluates the effectiveness and cost of slope mitigation methods under saturated conditions, considering stability, constructability, site constraints, and cost. Through this work, he aims to contribute practical and reliable solutions for safer and more resilient transportation infrastructure.

ABSTRACT

Traffic flow modeling provides a framework for analyzing the interactions of vehicles on roadways, enabling engineers to optimize traffic signal control and improve overall transportation efficiency. Traditionally, traffic flow models are categorized into three main classes: microscopic, mesoscopic, and macroscopic. Microscopic models focus on the individual interactions between vehicles, often represented through leader–follower dynamics, while macroscopic models describe traffic as a continuum characterized by aggregate quantities such as density and velocity. Mesoscopic approaches bridge these two perspectives but remain less commonly applied in large-scale predictive models.

The present work develops a unified model that links microscopic vehicle interactions with macroscopic traffic behavior through fluid mechanics principles. In this formulation, microscopic traffic dynamics are interpreted through an incompressible-flow analogy. This emphasizes local vehicle spacing and leader–follower velocity relationships, using vehicle trajectory data to generate source terms, that will be incorporated in the macroscopic momentum equations. Conversely, macroscopic traffic flow is treated as a compressible fluid capable of capturing variations in density, wave propagation, and shock formation associated with congestion. Coupling the microscopic and macroscopic traffic flow model will lead to the development of a hyperbolic–relaxation system capable of reproducing key traffic phenomena, including stop-and-go waves, nonlinear instability growth, and non-equilibrium traffic states.

By unifying incompressible and compressible flow analogies within the model, this research establishes a systematic pathway from vehicle-level interactions to continuum-scale traffic dynamics. The resulting model aims to improve the predictive capability of traffic flow models and provides new tools for transportation analysis and traffic control strategies.

PRESENTER

Oluwaseun Taiwo

Oluwaseun Ajadi is a Ph.D. student in the Department of Mechanical Engineering at the University of Kansas, where he also serves as a Graduate Teaching Assistant. He earned his bachelor's degree in mechanical engineering from Covenant University. His research focuses on applying principles of fluid dynamics and thermodynamics to traffic flow modeling, with a particular emphasis on developing macroscopic traffic models that incorporate microscopic driver behavior. His work aims to improve the prediction and understanding of complex traffic phenomena, including traffic congestion, shock waves, and stop-and-go traffic.

ABSTRACT

Reinforcement Learning (RL) has emerged as a promising paradigm for autonomous driving (AD) due to its ability to learn policies through dynamic interactions with complex driving environments. Recent advancements in RL-based AD have moved beyond standalone RL algorithms towards a diverse set of extended and hybrid learning paradigms, accompanied by specific design choices across driving tasks. However, existing works have yet to categorize and analyze how these paradigms shape design decisions and performance trade-offs across the AD stack. This paper aims to address this gap by providing a systematic understanding of the most recent development of the field through the joint lenses of driving tasks and learning paradigms.

We organize the literature primarily through the lens of learning paradigms while grounding each paradigm within functional driving layers, enabling an explicit mapping between the shortcomings of traditional RL in the choice of paradigm, design components, and algorithmic solutions. Through an indepth review of works published between 2023 − 2025, we analyze how major paradigms address the key limitations of vanilla deep RL, discuss open challenges as new hybrid paradigms arise, and motivate the RL-based AD outlook. Beyond synthesizing the emerging trend, this paper identifies recurring design patterns, paradigm-specific evaluation practices, and underexplored directions, offering a structured reference for researchers, practitioners, and new entrants to the field.

PRESENTER

Nhat Ha Nguyen

Nhat Ha is a PhD student in Computer Science at the School of Computing at Wichita State University. Her research focuses on developing safe and interpretable intelligent systems for autonomous driving, with particular interests in reinforcement learning, computer vision, and vision-language models. Her recent work explores pedestrian-aware autonomous driving, risk-sensitive decision making, temporal reasoning, and evidence-grounded evaluation of multimodal driving models. She is also interested in building benchmarks that assess whether AI systems have sufficient visual evidence to make reliable decisions rather than simply producing confident answers. Research Interests Reinforcement Learning Deep Learning Computer Vision Vision-Language Models Connected and Autonomous Vehicles Safe and Interpretable AI

ABSTRACT

Freight operations generate vast streams of location-based data that, if properly analyzed, support efficient freight logistics planning, tour-based modeling, supply chain optimization, and infrastructure development. While individual carriers can track their own trucks, state agencies or research institutes need insight into collective truck activity. Because agencies rely on anonymized probe data from commercial vendors, these products depend on vendor algorithms to infer trips, stops, and activity types. Understanding truck stop purposes is challenging when only anonymized GPS data is available. This study develops a scalable GPS-only rule-based framework to identify truck stop purposes that support logistics analysis without relying on external datasets. The resulting stop purpose classification method could also enhance vendor algorithms and improve freight activity products for agencies and industry. The proposed method categorizes stops into four key types: “Pickup/Delivery,” “Short Rest (&lt; 2 hours),” “Long Rest (&gt; 4 hours),” and “Staging,” while ensuring compatibility across both urban and non-urban environments. One-at-a-time sensitivity analysis identified key parameters influencing classification performance, leading to an optimized model with an 85% overall accuracy. The highest accuracy (94.5%) was seen for “Short Rest” classified stops. The model achieved 89.4% accuracy in non-urban areas and 81.1% in urban areas, demonstrating generalization across geographic contexts. The results demonstrate the algorithm’s effectiveness in distinguishing various stop purposes while highlighting challenges in differentiating “Pickup/Delivery” from “Short Rest” stops due to overlapping characteristics. The proposed approach enables improved representation of freight tours and facility interactions, providing actionable insights for logistics system analysis and infrastructure planning

PRESENTER

Mehdi Zolali

Mehdi Zolali is a Ph.D. candidate and Graduate Assistant in the Department of Civil Engineering at the University of Arkansas. His research focuses on freight transportation, truck GPS trajectory analysis, and machine learning applications for transportation systems. He is affiliated with the Freight Data Lab, where he develops data-driven methods for detecting truck stops and analyzing freight movement patterns using large-scale GPS data. His work integrates transportation engineering, spatial analytics, and artificial intelligence to improve the understanding of commercial vehicle operations. Mehdi has presented his research at national conferences and has contributed to studies published in transportation journals. His current research interests include freight data analytics, machine learning, transportation network analysis, and intelligent transportation systems.

ABSTRACT

Variable Speed Sign (VSS) systems are increasingly deployed by state DOTs to advise lower speeds during adverse weather, yet driver compliance with advisory (non-regulatory) recommendations remains poorly characterized in field conditions. The Nebraska Department of Transportation has installed a corridor-wide VSS system along I-80, with signs spaced approximately every eight miles. Because the displayed speeds are advisory rather than enforceable, characterizing driver response during winter weather is critical for evaluating system effectiveness and refining future operations. This study presents a paired temporal and spatial analysis of driver compliance during three winter 2025–2026 weather events (November 29, December 13, and February 19–20). Probe-vehicle speeds from the National Performance Management Research Data Set (NPMRDS), reported at 5-minute resolution on TMC segments, were merged with VSS activation logs and segment geometry. A temporal pipeline compares driver speeds before and after each activation at the same location, using non-overlapping time slices to isolate the advisory effect; a spatial pipeline compares upstream and downstream speeds within active advisory windows.

PRESENTER

Amir Sadeghnejad

Amir Sadeghnejad is a master’s student in Transportation Engineering at the University of Nebraska–Lincoln, where he works as a Graduate Research Assistant on projects funded by the Nebraska Department of Transportation. His research focuses on roadway safety, traffic operations, variable speed sign systems, and methods for estimating Annual Average Daily Traffic. He is also developing an Excel/VBA-based tool to improve traffic-count analysis and AADT estimation. Amir holds a bachelor’s degree in Civil Engineering from Iran University of Science and Technology, where he completed a comprehensive roadway design project. His technical experience includes OpenRoads Designer, Civil 3D, MicroStation, Python, R, Excel, and VBA. He is particularly interested in highway design, transportation data analysis, roadway safety, and applying engineering tools to improve transportation systems.