Center for Connected Multimodal Mobility

Building Smarter Cities via Intelligent Asset Management: South Carolina Case Study using IBM Maximo Application

Lead Principal Investigator – Paul Ziehl, University of South Carolina

Co-Principal Investigator(s) – Gurcan Comert, Benedict College, Nathan Huynh, University of South Carolina

September 2021 – November 2023

Project Status – Completed

Funding Amount – UTC $69,214, USC – $25,261, Benedict College – $6,668

Total Funding – $101,143

Sponsoring Orgs – OST-R, University of South Carolina, Benedict College

Description: Bridges serve as vital hubs in the national economy, facilitating the movement of goods and vehicles. South Carolina has nearly 9,455 bridges, and 8.7 percent are currently classified as structurally deficient [ARTBA, 2021]. A total of 89.4 percent of structurally deficient bridges are located on interstate highways and other critical roadways that connect major airports, ports, railroads, and truck terminals [ARTBA, 2021], posing a threat to the transportation of people and goods.

Intellectual Merit: The research goal is to develop a cost-effective approach to monitor the road conditions by cloud-based collaborative monitoring using in-vehicle smartphones which could be from any general public vehicle users.

Broader Impacts: This project aims to reduce the cost of road condition monitoring by providing a very cost-effective way with minimum investment of equipment and labors, significantly improve the safety of transportation systems, especially the multimodal connected and automated transportation systems, by providing timely needed road condition monitoring, and create a smartphone-based road condition dataset to benefit the research society.

A Statistical and Machine Learning Approach to Assess Contextual Complexity of the Driving Environment Using Autonomous Vehicle Data

Lead Principal Investigator – Jennifer Ogle, Clemson University

Co-Principal Investigator(s) – Gurcan Comert, Benedict College; Vijay Bendigeri, Consultant; Hudson Smith, Clemson University

September 2021  – May 2024

Project Status – Completed

Funding Amount – UTC $63,319, Clemson – $24991, Benedict College – $6,668

Total Funding – $94,978

Sponsoring Orgs – OST-R, Clemson University, Benedict College

Description: This project aims to understand the static and dynamic scene complexity from a driver’s perspective using the speed, density, and proximity of the objects around the vehicle and integrate machine learning to develop a Contextual Risk Factor (CRF) model to estimate the driving scene’s complexity and classify contextual risk. The output will be a heatmap that classifies the driving environment’s complexity into high, medium, and low categories. The researchers will use open-source LiDAR data collected by Waymo autonomous vehicles to estimate frame-by-frame road complexity considering dynamic traffic conditions. The LiDAR data provides rich real-world activity information around the vehicle, including stationary and non-stationary objects such as vehicles, pedestrians, and signs.

Intellectual Merit: This study considers the speed, density, and proximity of objects in the entire driving environment and within the driver’s cone of vision to develop a measure of the driving environment complexity. The authors plan to contrast the differences between the total environment complexity and the complexity within the driver UFOV.

Broader Impacts: Goal: Understand the dynamic scene complexity from a driver’s perspective and develop a heatmap using unsupervised clustering approaches that classifies the driving environment’s complexity.

Objectives:
• Measure contextual complexity and risk considering the dynamic components of the driving environment.
• Utilize data-rich LiDAR data collected by Waymo autonomous vehicles to reflect dynamic aspects of the environment.
• Apply unsupervised clustering methods to estimate the road environment’s complexity.

Real-time Decentralized Framework for Technology-Enabled Intermodal Freight Transport

Lead Principal Investigator – Nathan Huynh, University of South Carolina

Co-Principal Investigator(s) – William Ferrell, Clemson; Negash Begashaw, Benedict College; Gurcan Comert, Benedict College

September 2021 – October 2023

Project Status – Completed

Funding Amount – UTC $100,721.00, USC – $28,775, Clemson – 15,816, Benedict College – $9,609

Total Funding – $154,911

Sponsoring Orgs – OST-R, University of South Carolina, Clemson University, Benedict College

Description: In 2018, the U.S. transportation system moved approximately 51.0 million tons of freight daily valued at more than $51.6 billion, and this tonnage is forecast to increase at about 1.2% per year until 2045. Although the U.S. spends a bit less than the international average on logistics as a percentage of GDP, there still exist significant inefficiencies. For example, approximately 25% of truck miles traveled were with completely or nearly empty trailers and the remaining 75% of miles were 56.8% full trailers (Matthams, 2019). This level of inefficiency couples with only increasing volume would represent a significan challenge; however, the situation is worse because the nature of freight is changing due to the growth of eCommerce. Historically, products were frequently moved on pallets and delivered to stores. Today, this is being replaced by an increasingly large number of small packages being delivered to individual customers. The current inefficiency plus a growing volume of freight is having a significant, direct negative impact on energy use and associated environmental externalities. Needlessly consuming fuel (e.g., to reposition assets or moving partially loaded trucks) is both costly and detrimental to the environment. As volume increases, both the costs and the carbon footprint will increase as well. Moreover, the significant amount of variability caused by changing customer requirements and system disruptions mean freight transportation operates in a stochastic environment that adds to the challenge. There is a critical need for freight carriers to operate more efficiently and more effectively despite these many obstacles.

Intellectual Merit: This research project will fulfill a critical need in freight logistics by developing a hybrid centralized-decentralized framework in which multiple less-than-truckload (LTL) carriers operating in a stochastic and dynamic environment will simultaneously collaborate and complete.

Broader Impacts: The overarching goal of this project is to develop a real-time centralized-decentralized freight logistics framework to assess how carrier collaboration and trucks making decentralized decisions could reduce cost and logistics carbon footprint. The specific objectives of this project are:

1. Formulate a mathematical modeling approach to address the decisions to be made by the edge devices. This approach must address dynamic routing and jobs to outsource/acquire.
2. Create a framework for the central clearinghouse/auctioneer component that processes information quickly and effectively to run an expedited auction. Investigating the opportunity to incorporate Artificial Intelligence (AI) to achieve this is part of this objective.
3. Construct an approach to reduce large amounts of data that will be available to the edge devices into parameters that are required by the model discussed in #1. This objective will also seek to make use of AI.
4. Develop algorithms to solve proposed decentralized model(s) with a special emphasis on computational efficiency to allow the algorithms to run on edge devices.
5. Develop a micro-traffic simulation model to simulate realistic movements of trucks in a road network with typical recurrent and non-recurring congestion and disruptions.
6. Develop a real-time interface to enable communications between all edge devices.
7. Use developed real-time system to conduct numerical experiments to gain insight on system and operating factors that affect logistics cost and carbon footprint.

Securing Deep Learning against Adversarial Attacks for Connected and Autonomous Vehicles

Lead Principal Investigator – Pierluigi Pisu, Clemson Univ. I-CAR

Co-Principal Investigator(s) – Gurcan Comert, Benedict College; Negash Begashaw, Benedict College

September 2021 – October 2023

Project Status – Completed

Funding Amount – UTC $69,215, Clemson – $25,000, Benedict College – $9,609

Total Funding – $103,824

Sponsoring Orgs – OST-R, Clemson University, Benedict College

Description: The overall vision of this project is to develop a defense technique capable of making CAVs more resilient to adversarial attacks and therefore able to satisfy more stringent system’s safety and performance requirement. In order to achieve this vision, this project focuses on the following set of goals:

• Implement the state-of-the-art adversarial attack on a camera and Lidar fusion-based object detection algorithm in the physical world.
• Develop a defense technique for CAVs that utilizes camera and Lidar signals to mitigate physical adversarial attacks in the CAV perception module. This will enhance vehicle performance and improves the security and reliability of autonomous systems.
• Validate the proposed approach and the performance of the object detection unit on datasets with camera and Lidar signals as inputs.
• Evaluate and demonstrate the proposed technology on an autonomous F1/10 car testbed whose perception architecture mimics a CAV perception module.

Intellectual Merit: As CAV technology is developing fast and going to enter the market soon, the proposed research addresses the problem of improving the resilience of CAVs to the possibility of adversarial attacks aimed at affecting the performance of perception module of CAVs, therefore improving vehicle reliability and functional safety beyond currently adopted practices. We envision that our technique will play an important role in securing automated vehicles and thus, accelerating the spreading of CAVs. Expected outcomes of the project fall well within the C²M² research priority focus on artificial intelligence in multi-modal transportation cyber-physical systems.

Broader Impacts: The overarching of this project is to develop a technique to improve the trustworthiness of perception information used by CAVs. More specifically, this project proposes a defense mechanism against adversarial attacks performed on the 3D object detector in CAV perception module. The project will focus on an autonomous vehicle architecture with perception-planning-action pipeline⁴. The results of the perception module are going to be used in the planning module to execute motion planning task. In this depicted architecture, CAV faces the challenge of obtaining correct sensing information about surrounding environment including recognizing pedestrians and traffic signs. The proposed technique will also account for real-time computational constraints and tradeoff between accuracy and robustness. This objective will be achieved by developing a defense mechanism for CAVs to obtain trustworthy inputs from camera and Lidar by making CAVs perception module robust to adversarial inputs. The specific objectives of the proposed research consist in developing: (a) an approach to implement adversarial attacks on CAV sensor fusion system and (b) an intrinsically robust neural networks to make adversarial attacks less effective, which means to obtain correct recognition results even under adversarial attacks. The real-time evaluation of this strategy will be conducted using an autonomous F1/10 car testbed and the performance will be compared against the baseline model (non-robustified model).

Multimodal-AI based Roadway Hazard Identification and Warning using Onboard Smartphones with Cloud-based Fusion

Lead Principal Investigator – Yunyi Jia, Clemson Univ. I-CAR

Co-Principal Investigator(s) – Gurcan Comert, Benedict College

September 2021 – July 2023

Project Status – Completed

Funding Amount – UTC $51,215, Clemson – $16,000, Benedict College – $9,609

Total Funding – $76,824

Sponsoring Orgs – OST-R, Clemson Univ., Benedict College

Description: We have dealt with the problem of detecting pavement conditions using in-vehicle smartphone in our previous C2M2 project. In this proposal, with the suggestion of SC-DOT on their demands, we are going to leverage our existing achievements and achieve the following new objectives:

• Objective 1: Collect and annotate smartphone-based roadway hazards data.
• Objective 2: Investigate state-of-the-art machine learning based approaches to detect multiple types of roadway hazards and their level of threat.
• Objective 3: Investigate cloud-based fusion approach to fuse all roadway hazard detection results from different vehicles in the cloud in order to give a holistic, accurate and complete monitoring of roadway hazards and their threat levels.

Intellectual Merit: The merit of the project can be summarized as below

• Develop a very cost-effective approach to identify roadway hazards using in-vehicle smartphones of public vehicle users
• Develop a novel cloud-based collaborative roadway hazard monitoring approach with multi-modal-multi-output deep learning-based hazard detection and threat level estimation to deliver a holistic, accurate and complete monitoring of road conditions
• Create a smartphone-based roadway hazard dataset for training road hazard detection models

Broader Impacts: The major expected impacts are summarized as below

• Provide a very cost-effective way of identifying roadway hazards with minimum investment of equipment and labors.
• Significantly improve the safety of transportation systems, especially the multimodal connected and automated transportation systems, by providing timely needed roadway hazard information.
• Create a smartphone-based roadway hazard dataset to benefit the research society.

A Machine Learning-Assisted Framework for Determination of Performance Degradation Causes and Selection of Channel Switching Strategy in Vehicular Networks

Lead Principal Investigator – Chin Tser Huang, USC

Co-Principal Investigator(s) – Gurcan Comert, Benedict College; Pierluigi Pisu, Clemson Univ. I-CAR; Esmail Abuhdima, Benedict College

September 2021 – June 2023

Project Status – Completed

Funding Amount – UTC $94,215, USC, $25,000; Clemson – $12,500; Benedict College – $10,109

Total Funding – $141,844

Sponsoring Orgs – OST-R, USC, Clemson University, Benedict College

Description: This study aims to keep vehicle connectivity reliable and persistent by designing an AI-assisted framework for automated, adaptive channel switching when severe performance degradation is detected (see Figure 1). We will achieve our goals via developing the following:

• A testbed environment for generating and collecting data for training and testing purposes. We will identify all performance degradation causes that are of interest, and deploy transceivers in our vehicles and test under real or synthetic conditions to collect and record data.
• A set of machine learning based models for detecting and classifying performance degradation conditions. We will construct machine learning based models from a variety of supervised machine learning algorithms for detecting performance degradation and classifying it. We will use the data collected from the testbed environment to train and test the models.
• An automated AI-assisted framework for integrating determination of performance degradation cause and selection of best channel switching strategy. We will study how the transceivers can discover and agree on the available channels, and then investigate and establish the relationship between the cause of performance degradation and the channel switching strategy that can best mitigate the cause. An appropriate threshold for performance drop will be found and used as the triggering mechanism of channel switching.

Intellectual Merit: In this study, we propose to construct a machine learning based model to detect and determine the most likely cause of performance degradation. Currently, there is a lack of effective mechanism which can quickly switch to a secondary channel when the performance of vehicular communication drops severely to guarantee the quality of performance. We aim to develop an automated framework which will make use of the learned cause to adaptively select the optimal channel switching strategy to mitigate the performance drop. Our ongoing project shows that there are different patterns of performance degradation which can be attributed to different causes, and researchers have designed some channel switching strategies. Our approach will apply AI techniques to automate the procedure to ensure the reliability of vehicular connectivity.

Broader Impacts:

• Machine learning models for detecting performance degradation and determining the most likely cause will be constructed.
• Adaptive framework which integrates machine learning model and automated channel switching will be developed and applied in vehicular communication applications.

We anticipate that our models will be a significant step towards modeling more uncertainty of various new causes that will lead to performance degradation in vehicular network communications. Thus, our models would likely be adopted by many researchers and industry to make their applications robust and adaptive to performance degradations.

A Cloud-based Quantum Artificial Intelligence-supported Truck Platooning Strategy for Safety and Operational Performance

Lead Principal Investigator – Mashrur “Ronnie” Chowdhury, Clemson University

Co-Principal Investigator(s) – Gurcan Comert, Benedict College; Dimitra Michalaka, The Citadel; Judith Mwakalonge, SCSU; Nathan Huynh, USC; William J. Davis, The Citadel; Kweku Brown, The Citadel

September 2021 – Active

Project Status – Report in progress

Funding Amount – UTC $194,955, Clemson – $16,000; Benedict College – $9,609; The Citadel, $30,532; SCSU, $16,309; USC, $28,775

Total Funding – $296,262

Sponsoring Orgs – OST-R, Clemson University, Benedict College, The Citadel, SCSU. USC

Description: Our research aims to develop a new Q-AI driven technology that will address the existing knowledge gap identified by existing literature of fully quantifying the impact of freight operation by integrating freight pathways, powertrain technologies, the total cost of ownership, infrastructure, and platooning technology. To achieve this goal, the research objectives are to, 1) develop cloud-assisted truck platooning models considering different truck powertrain technologies (Obj.-1), 2) develop cloud computing strategies to satisfy real-time computing requirements of structured and unstructured data from heterogeneous data sources (Obj.-2); 3) study the impacts of cloud-supported truck platoons, in terms of safety, operational efficiency and energy consumption,  along freight corridors in a simulation tool (Obj.-3); 4) validate the improved predictive analytics including Q-AI strategies in actual fleet trials for safety and operational impacts(Obj.-4), and 5) evaluate the cost-effectiveness and other impacts of proposed technology with the baseline technology with roadside units-supported cloud servers.

Intellectual Merit: To maximize the expected safety and operational benefits of truck platooning, this proposal intends to develop a cloud infrastructure supported platooning algorithm to assist truck platooning to minimize delay, reduce energy consumption and improve safety and demonstrate the application in the real world. Cloud infrastructure will provide a seamless, on-demand data storing, application hosting, and execution platform for truck platooning application. 

Broader Impacts: Reliance on in-vehicle computational devices for truck platooning, as considered in the existing studies, will increase the computational burden for each vehicle. To reduce the overreliance on the in-vehicle computing nodes and enable a predictive analytics-based truck platooning for a corridor to ensure safety and operational improvement, a cloud-based truck platooning framework will be developed in this research. Both connected trucks and automated trucks will be considered in this study. This research focuses on predictive analytics using quantum artificial intelligence (Q-AI). An earlier study discussed using Q-AI to enhance learning efficiency, learning capacity, and run-time improvements. The focus of this study is to develop predictive Q-AI algorithms for cloud-based, safe, and efficient truck platooning using high volume and heterogeneous data from multiple diverse sources with the capability of scaling up. This study will also evaluate the efficacy of platooning for safety and operational performance in the real world in a test track in Greenville, South Carolina.