1,721,099 research outputs found
Transfusor: Transformer diffusor for controllable human-like vehicle lane-changing trajectory generation
No full textThe importance of autonomous driving systems (ADS) and advanced driver assistance systems (ADAS) testing continues to pose a suject of vital importance. For doing this, virtual simulation testing (VST) has become a prominent approach due to its advantages of fast execution, low cost, and high repeatability. However, the success of such simulation-based experiments heavily relies on the realism of the testing scenarios. To enhance realism in such experiments, this paper introduces the "Transfusor" model, which leverages cutting-edge deep learning generative technologies, namely the diffusion model and Transformer. The primary objective of the Transfusor model is to generate highly realistic and controllable human-like lane-changing trajectories to drive the background traffic in highway scenarios. The results from extensive experiments demonstrate that the proposed model has the potential to effectively learn the spatiotemporal characteristics of lane-changing behaviors and successfully generate trajectories that closely mimic real-world human driving. With its ability to produce human-like lane-changing trajectories, this model contributes to creating more flexible and high-fidelity testing scenarios in VSTs, ultimately leading to safer and more reliable ADS and ADAS.
Dynamic urban traffic rerouting with fog-cloud reinforcement learning
No full textDynamic rerouting has been touted as a solution for urban traffic congestion. However, its implementation is stymied by the complexity of urban traffic. To address this, recent studies suggest the efficacy of novel technologies like fog computing and deep reinforcement learning. However, there exist significant challenges in this regard: (1) sorting massive amounts of data associated with large urban networks, (2) large action space that hinders learning efficiency, (3) impairment of rerouting efficacy due to overreliance on regional/local information, and (4) the issue of congestion shifting. To overcome these challenges, this paper presents a novel two-step approach that integrates graph attention Q network (GAQ) with entropy-balanced k shortest path (EBkSP) via fog-cloud information framework to carry out vehicle rerouting in dynamic traffic environments typical of large cities. The first challenge is addressed by GAQ that uses attention mechanism to evaluate traffic and assign region indices based on the relative importance of information. The second challenge is addressed using fog nodes that reduce the action space, and the third is addressed using fog computing combined with cloud computing to facilitate the computation of global optimal information in a centralized planning decentralized execution pattern. The fourth challenge is addressed using EBkSP, which identifies vehicle optimal routes based on route popularity and vehicle priority. After conducting a numerical experiment involving two other cutting-edge vehicle rerouting models as baselines, it was established that the proposed model delivers results that are 14%-54% superior from perspectives of the mean travel speed and congestion. The superior efficacy of the proposed learning-based (compared to rule-based) models is pronounced when rerouting ratios are low. As the rerouting ratio increases, both the learning-based and rule-based model outcomes exhibit reduced likelihoods of severe congestion. However, the learning-based model consistently outperforms the rule-based model across all scenarios. The proposed model can effectively reroute vehicles under different situations of rerouting ratio and total count of vehicles and can help improve traffic mobility and reduce congestion in urban areas. It can be implemented easily by urban road agencies.
Why did the AI make that decision? Towards an explainable artificial intelligence (XAI) for autonomous driving systems
No full textUser trust has been identified as a critical issue that is pivotal to the success of autonomous vehicle (AV) operations where artificial intelligence (AI) is widely adopted. For such integrated AI-based driving systems, one promising way of building user trust is through the concept of explainable artificial intelligence (XAI) which requires the AI system to provide the user with the explanations behind each decision it makes. Motivated by both the need to enhance user trust and the promise of novel XAI technology in addressing such need, this paper seeks to enhance trustworthiness in autonomous driving systems through the development of explainable Deep Learning (DL) models. First, the paper casts the decision-making process of the AV system not as a classification task (which is the traditional process) but rather as an image-based language generation (image captioning) task. As such, the proposed approach makes driving decisions by first generating textual descriptions of the driving scenarios, which serve as explanations that humans can understand. To this end, a novel multi-modal DL architecture is proposed to jointly model the correlation between an image (driving scenario) and language (descriptions). It adopts a fully Transformer-based structure and therefore has the potential to perform global attention and imitate effectively, the learning processes of human drivers. The results suggest that the proposed model can and does generate legal and meaningful sentences to describe a given driving scenario, and subsequently to correctly generate appropriate driving decisions in autonomous vehicles (AVs). It is also observed that the proposed model significantly outperforms multiple baseline models in terms of generating both explanations and driving actions. From the end user’s perspective, the proposed model can be beneficial in enhancing user trust because it provides the rationale behind an AV’s actions. From the AV developer’s perspective, the explanations from this explainable system could serve as a “debugging” tool to detect potential weaknesses in the existing system and identify specific directions for improvement.
Econometric models for pavement routine maintenance expenditure
No full textRoutine maintenance expenditures make up a significant fraction of the overall life-cycle costs incurred by highway agencies, and thus constitute a key input in cost evaluation. Most life-cycle cost analyses have proceeded only with very rough approximations of average annual maintenance expenditure due to difficulty in acquiring data. In addressing this issue, this thesis uses data from in-house maintenance records and other data sources to develop a cohesive and comprehensive dataset covering state highway pavement sections in Indiana, a state located in the wet-freeze climatic region in the United States. To assist in budgeting and life-cycle cost analysis, this thesis developed annual maintenance expenditure models using an array of statistical and econometric techniques, including ordinary least square, tobit, panel, and two-stage regression. This thesis identifies a number of explanatory variables that significantly influence maintenance expenditures and examines the sensitivity of the response to each of these variables. Specifically, the geographic region, pavement segment length, and age were found to be significant indicators annual routine pavement maintenance expenditure
A stochastic multi-criteria assessment of security of transportation assets
Get PDFTransportation project evaluation and prioritization use traditional performance measures including travel time, safety, user costs, economic efficiency, and environmental quality. The project impacts in terms of enhancing the infrastructure resilience or mitigating the consequences of infrastructure damage in the event of disaster occurrence are rarely considered in project evaluation. This dissertation presents a methodology to address this issue so that in evaluating and prioritizing investments, infrastructure with low security can receive the attention they deserve. Secondly, the methodology can be used for evaluating and prioritizing candidate investments dedicated specifically to security enhancement. In defining security as a function of threat likelihood, asset resilience and damage consequences, this dissertation uses security-related considerations in investment prioritization thus adding further robustness in traditional evaluations. As this leads to an increase in the number of performance criteria in the evaluation, the dissertation adopts a multiple-criteria analysis approach. The methodology quantifies the overall security level for an infrastructure in terms of the threats it faces, its resilience to damage, and the consequences in the event of the infrastructure damage. The dissertation demonstrates that it is feasible to develop a security-related measure that can be used as a performance criterion in the evaluation of general transportation projects or projects dedicated specifically towards security improvement. Through a case study, the dissertation applies the methodology by measuring the risk (and hence, security) of each for bridge infrastructure in Indiana. The method was also fuzzified and a Monte Carlo simulation was run to account for unknown data and uncertainty. On the basis of the multiple types of impacts including risk impacts such as the increase in security due to each candidate investment, this dissertation shows how to prioritize security investments across the multiple infrastructure assets using multiple-criteria analysis
Lane Management in the Era of Connected and Autonomous Vehicles Considering Sustainability
No full textThe last century has witnessed increased urban sprawl, motorization, and the attendant problems of congestion, safety, and emissions associated with current-day transportation systems. Contemporary literature suggests that emerging transportation technologies, including vehicle autonomy and connectivity, offer great promise in addressing these adversities. As such, highway agencies seek guidance on infrastructure preparations for connected and automated vehicle (CAV) operations. A key area of such preparations is the management of lanes to serve CAVs and humandriven vehicles (HDVs), including the deployment of dedicated lanes for CAVs. There is a need to address the demand and supply perspectives of CAV preparations. On the demand side, agencies need to model the trends and uncertainties of CAV market penetration and level of autonomy during the CAV transition period. On the supply side, agencies need to schedule the CAV-related roadway infrastructure in a way that progressively addresses the growing demand.In addressing these research questions, this dissertation first carries out an economicsbased lane allocation for CAVs and HDVs in a highway corridor by determining the optimum number of CAVLs by minimizing road user cost. Next, the dissertation carries out such allocation considering the environment (community emissions cost). Third, the dissertation addresses elements of social and economic sustainability using a CAV-enabled tradable credit scheme that minimizes user travel time subject to social equity constraints. Further, this dissertation provides guidance on how CAV-dedicated lanes, in conjunction with market-based tradable travel credits, could enable the road agency to achieve maximum efficiency of the existing road infrastructure in the CAV transition period. The study framework can serve as a valuable decision-support tool for road agencies in their long-term planning and budgeting in anticipation of the CAV transition period. The key outcome of the framework is an optimal schedule for deploying CAV-dedicated lanes over a given analysis period of several decades in a manner commensurate with CAV demand projections and sustainability-related objectives and constraints
Deterioration modeling of highway bridge components using deterministic and stochastic methods
No full textProper timing of bridge repair or reconstruction is critical for the bridge management functions including long-term planning and budgeting, and hinges heavily upon the reliability of bridge deterioration prediction. Bridge deterioration models can be used not only for making such predictions but also to identify the factors that affect bridge component deterioration, and to measure the strength of these factors. The current literature lacks adequate and comprehensive assessment of bridge component deterioration using recent data that accounts for a wide range of potential explanatory factors including service type, design type, and climate. In addressing this lacuna in the literature, this thesis carefully assembled a comprehensive bridge dataset using data from the National Bridge Inventory and the National Oceanic and Atmospheric Administration, developed a technique for data quality/integrity enhancement, and estimated statistical models for bridge components for each of several bridge families. The bridge families were created based on their functional classification and rehabilitation history. For each component (deck, superstructure and substructure) within each family, deterministic and stochastic models were developed. The deterministic models used ordinary least squares estimation, while the stochastic models used an ordered probit specification. The models also threw more light on the direction and strengths of influence of the explanatory factors of bridge component deterioration, including service type, design type, material type, and climate. Age was consistently found to be the most influential factor. The study results are useful for the various tasks associated with bridge management including maintenance and rehabilitation programming and budgeting, cost allocation and bridge asset valuation
Mixed linear modeling techniques for enhancing pavement performance predictions
No full textThe use of appropriate advanced modeling techniques for predicting the performance of pavements that have received rehabilitation treatments may reap substantial benefits to a Pavement Management System (PMS). If the modeling technique is appropriately chosen on the basis of practicality, precision, the intended use of the model, and the nature of the pavement data, its applicability to PMS can be enhanced greatly. Pavement rehabilitation data typically constitutes of repeated measurements that form an unbalanced three-level nested structure, which makes the analysis quite challenging. This thesis proposes an enhanced methodological framework for pavement rehabilitation treatment analysis that uses mixed linear modeling techniques. Mixed models constitute a statistical technique that includes both fixed effects and random effects. The proposed framework is demonstrated using data from the Indiana Interstate network. In applying the developed framework, agencies can not only statistically quantify the post-rehabilitation performance of pavements, but also develop estimates and ranges of treatment service lives and thus update or refine the treatment service lives that are currently published in their pavement design or preservation manuals. These procedures are demonstrated analytically using a case study. The proposed framework can also be used by highway agencies as part of their network-level needs assessment because it offers a more reliable estimation of future physical and fiscal needs, as shown in the case study presented in this thesis
Development of framework for statewide vehicle miles traveled (VMT) estimation
Get PDFVehicle Miles Traveled (VMT) is a critical performance measure that is used extensively in highway transportation management for financial analysis, resource allocation, impact assessments, and reporting to oversight agencies. As highway revenue from fuel taxes continues to plummet and user-based taxes such as VMT fees become increasingly attractive, consistent and reliable VMT estimates have become critical for highway funding evaluation and administration. At the present time, there are several methods for VMT estimation that typically yield estimates that are inconsistent or inaccurate. This thesis presents alternative techniques for VMT estimation in the state of Indiana at the project, regional, and network levels for confirming or estimating the levels and distribution of vehicular travel at the present time as well as at any specified future time. The present research also developed a benchmark method (segment-level using traffic counts) for VMT estimation and shows how the estimates from the other different methods can be calibrated to mitigate the inconsistencies in statewide VMT estimation across the different methods. The early tasks of the research, which included a literature review and survey of VMT-data stakeholders, helped streamline the research effort, categorize the different techniques for VMT estimation and identify their limitations, and identified the preferred outputs of any platform for VMT estimation. The core outcome of this thesis is a comprehensive framework for estimating the VMT contributed by each vehicle class for the state’s entire road network. This framework estimates statewide VMT by using the segment length, traffic volume, and distance, for the primary highway systems of state routes (interstates, US and state roads) and local routes (city streets and county roads). Local route VMTs were studied in-depth because of their historical underrepresentation in VMT studies, the low accuracy of past estimating methods, and the local road’s significant share of the total road inventory. For the state road VMT estimation, a comprehensive database was developed that facilitates extensive aggregations of VMT by geographical scope, route, functional class, and vehicle class. For the local-route VMT estimation, a sample of counties of different spatial locations and degrees of urbanization were used. Analytical techniques and tools, including cluster analysis, geographic information systems (GIS), and spatial interpolation techniques were used to expand the VMT estimates from the local road sample to the population of all counties in the state. The results indicate that there is a -21% (underestimate) to +8 % (overestimate) in the results from the various VMT estimation methods, as compared to the benchmark method (segment-level VMT estimation) developed in this research. The technique developed in this research for reconciling these different VMT estimates was validated using the estimate from the benchmark method as a basis. The implementation platform developed in this research was designed to produce outcomes that address the VMT data needs of a state highway agency and other stakeholders, and could be enhanced in the future as and when data become available. The deliverables from this research are expected to have far-reaching impacts on the various functional areas of highway management and administration, the evaluation of a VMT fee as an alternative or complement to the fuel tax for highway revenue, and the generation of required reports to federal oversight agencies
- …
