No description available.

Budgets for transportation improvements are limited so it is important for governments to focus on improving locations most in need of safety funding. The objective of the Two-Output Model for Safety (TOMS) is to provide the Utah Department of Transportation (UDOT) a reliable method to prioritize safety improvements on state-owned roadways among the different regions. This research will improve the existing Crash Analysis Methodology for Segments (CAMS) and Intersection Safety Analysis Methodology (ISAM) being used to analyze crashes on Utah roadways. The scope of this project is improving on the existing CAMS and ISAM to work together within R, to incorporate segment and intersection severity in safety hot spot analysis, to develop overall severity distributions, and to develop limited recommendations and conclusions related to the research. TOMS uses UDOT data to create a statistical input. Each segment is homogenous with respect to five variables: average annual daily traffic, functional class, number of through lanes, speed limit, and urban code. Intersections are provided as a separate dataset. In the statistical analyses performed on the data, five years of crash data (2016-2020) are used to determine a weighted risk score for segments and intersections of similar characteristics. Those segments or intersections with excess weighted risk scores are designated as crash hot spots. Two-page technical reports with road characteristics and crash data are created for the top 10 hot spots for segments and intersections in Utah. The reports are sent to UDOT where region engineers may review and determine which locations might be addressed.

The standard method for assessing traffic safety is to use the predictive method outlined in the Highway Safety Manual (HSM). This method is site-level, data-intensive, and does not account for interactions between sites, making it difficult to assess larger areas. This dissertation develops a corridor-level approach to traffic safety which uses less data than the HSM predictive method and views roadways holistically rather than combinations of individual, independent sites. First, a corridor definition is developed and applied to 10 urban Florida counties with a history of many crashes, resulting in the identification of 1,048 corridors. These corridors were primarily defined using context classification and lane count, with additional considerations for data availability and minimum length. From 2017–2021, these corridors experienced 459,603 unique crashes. After preliminary modeling and scope refinement, 559 corridors received supplemental data collection. Between the two datasets, a total of 11 models were developed using either negative binomial (NB) or random forest (RF) regression. NB models can be used for network screening purposes or identifying the impacts of potential safety improvements, while RF models can be used to identify variables important to the accuracy of the prediction. Potential safety improvements identified from the NB models include increasing proactive law enforcement patrols for dangerous driving behaviors and installing corridor lighting in corridors without lighting. While both NB and RF models were accurate, NB models were recommended due to resulting in a definite equation and overdispersion parameter that could be used with the empirical Bayes (EB) method to improve prediction accuracy. Overall, the corridor-level NB models outperformed the HSM models in terms of accuracy and statistical reliability. Using a corridor-level approach can help agencies quickly network screen their systems to identify high-risk corridors in need of safety improvements or supplement site-level analyses.

Improvements to highway safety have become a high priority for highway authorities due to increasing public awareness and concern of the high social and economic costs of accidents. However, satisfying this priority in an environment of limited budgets is difficult. It is therefore important to ensure that the funding available for highway safety improvements is efficiently utilized. In attempt to maximize the overall highway safety benefits, highway professionals usually invoke an optimization process. The objective of this thesis study is to develop a model for the selection of appropriate improvements on a set of black spots which will provide the maximum reduction in the expected number of accidents (total return), subject to the constraint that the amount of money needed for the implementation of these improvements does not exceed the available budget. For this purpose, a computer program, BSAP (Black Spot Analysis Program) is developed. BSAP is comprised of two separate, but integrated programs: the User Interface Program (UIP) and the Main Analysis Program (MAP). The MAP is coded in MATLAB and contains the optimization procedure itself and performs all the necessary calculations by using a Binary Integer Optimization model. The UIP, coded in VISUAL BASIC, was used for monitoring the menu for efficient data preparation and providing a user-friendly environment.

This study explores the application of multi-agent reinforcement learning (MARL) to enhance the decision-making, safety, and passenger comfort of Autonomous Vehicles (AVs)at uncontrolled intersections. The research aims to assess the potential of MARL in modeling multiple agents interacting within a shared environment, reflecting real-world situations where AVs interact with multiple actors. The findings suggest that AVs trained using aMARL approach with global experiences can better navigate intersection scenarios than AVs trained on local (individual) experiences. This capability is a critical precursor to achieving Level 5 autonomy, where vehicles are expected to manage all aspects of the driving task under all conditions. The research contributes to the ongoing discourse on enhancing autonomous vehicle technology through multi-agent reinforcement learning and informs the development of sophisticated training methodologies for autonomous driving.