Electrical and Computer Engineering

Permanent URI for this collectionhttps://hdl.handle.net/10323/11889

Browse

Now showing 1 - 20 of 59

A Complete Deep Learning-Based Driver Monitoring System with Low Cost
(2026-01-01) Liu, Bing; Ganesan, Subramaniam; Deng, Xiaodong; Liu, Anyi; Alawneh, Shadi
This research provides a solution to build a deep learning-based driving monitoring system (DDMS) to verify the driver’s ID, estimate the driver’s status and the vehicle’s occupancy status. In our DDMS, we built and trained three deep learning neural networks: object detection neural network (ODNN), facial recognition neural network (FRNN) and facial landmark detection neural network (FLDNN). We built software to drive a camera to capture images in the vehicle and integrate these three neural networks to build our DDMS. The pictures of the driver seat, passenger seat and back seat are sent to ODNN. ODNN detects common objects in vehicles such as humans, faces, dogs, and so on; the driver’s facial images are sent to FRNN. Our DDMS uses the facial recognition result to verify if the person is authorized to drive; the driver’s face images are also sent to FLDNN. FLDNN detects facial keypoints including eyebrows, eyes, the nose, lips, and the chin continuously. Our DDMS estimates driver’s state such as activeness, drowsiness, distractedness and so on. The movement of lips detected by FLDNN and other detected objects by ODNN near the driver’s face such as cell phone are used to estimate if the driver is speaking on the phone. Public datasets that meet our requirements to train our ODNN do not exist to the best of our knowledge. We built an Auto-Labeling Model (ALM) to expand an existing dataset by assigning labels to the data automatically. Quantizing deep learning models from floating point to 8-bit or lower-bit scaled integer data types is required to reduce the cost of using DDMS. Quantization is applied to convert the parameters in ODNN, FRNN and FLDNN from floating point numbers to integer numbers. The contribution in our study includes building and training an ALM to create a new dataset for ODNN; building and training ODNN, FRNN and FLDNN; developing software to drive the camera and integrating ODNN, FRNN and FLDNN together to provide vehicle’s occupancy status, face ID and driver status estimation results; quantizing ODNN, FRNN and FLDNN.
Unified Nonlinear Model Predictive Path-Following Control with Collision Avoidance for Robotic Systems
(2026-01-01) Dhansri, Naren Reddy; Das, Manohar; Qu, Guangzhi; Li, Li
Robotic systems operating in constrained environments must often follow prescribed geometric paths while respecting nonlinear dynamics, actuator limits, state constraints, and obstacle-clearance requirements. Conventional trajectory tracking prescribes both path and timing, which can become restrictive when the motion is dynamically infeasible or when obstacle avoidance requires temporary deviation from the prescribed path. This dissertation develops a unified nonlinear Model Predictive Path-Following Control (MPPFC) framework with collision avoidance, in which path following, path progression, dynamics, constraints, and obstacle avoidance are jointly optimized within a single Model Predictive Control (MPC) problem. The central contribution is a complete computational framework that makes the unified collision-aware MPPFC formulation practical by developing analytical collision-constraint derivatives and a tailored primal–dual interior-point solver for the resulting nonlinear control problem. Progression along the prescribed path is optimized online rather than imposed through a fixed time parameterization, allowing adaptation when actuator limits, dynamic constraints, or obstacle-clearance requirements become active. Smooth composite Bezier paths provide analytical path derivatives, while differentiable minimum-distance collision constraints provide analytical gradients and Hessians through closest-point sensitivity analysis and articulated kinematic propagation. The resulting nonlinear optimal-control problem is solved using a second order primal–dual interior-point method tailored to the structured prediction horizon formulation of MPPFC. The solver incorporates analytical derivatives, slack and relaxation variables, barrier terms, scaling, regularization, warm starting, and sparse Newton-system solution. By preserving the stage-wise temporal structure, the method efficiently handles path-following objectives, nonlinear dynamics, constraints, and collision-avoidance inequalities online. The proposed methods are evaluated on two robotic systems: an articulated tractor–trailer navigating around static and dynamic obstacles, and an underactuated planar log-arm robot performing constrained path following. These studies show that adaptive path progression, obstacle avoidance, constraint satisfaction, and feasible control generation can be handled within a unified predictive control framework for nonlinear systems subject to actuator limits, obstacle interactions, changing clearance constraints, and limited actuation, while preserving a computational structure suitable for future real-time embedded implementation.
Can Network Security Protocol of Autonomous Vehicle Based on FPGA in the Loop
(2026-01-01) Lagnf, Farag Mohamed E; Ganesan, Subramaniam; Liu, Anyi; Alawneh, Shadi; Deng, Xiaodong
In Software-Defined Vehicles (SDVs) the transition towards zonal topologies decreases the attack surface and improves the security requirements of in-vehicle communication networks such as CAN FD. This dissertation introduces a lightweight FPGA-based security frame designed to mitigate replay attacks and spoofing threats with a deterministic synchronized Freshness Value (FV) derived from a non linear function. The FV is transformed into a cryptographically balanced bitstream and integrated with AES 128 Encryption and lightweight authentication methods to make sure secrecy, integrity, and freshness verification with low computing burden. The proposed architecture was executed and verified by an FPGA-in-the-Loop (FIL) codesign approach on a Xilinx Arty A7 platform. MATLAB and VHDL integration have been utilized. With loss of 30 % of CAN FD messages, experimental findings indicated that precise synchronization has achieved between transmitter (Tx) and receiver Rx) at 100 MHz, with observed latencies of 170 ns for Tx and 160 ns for Rx, resulting in an overall throughput of roughly 123 bps. encryption, ensures freshness and reduces replay attacks, thereby overcoming the limitations of conventional freshness mechanisms. In order to accomplish this, we suggest a hardware architecture that is predicated on an FPGA (Arty 7A), which has been verified by our simulation results to provide high throughput and minimal payload. The real-time hardware implementation effectively limits replay attacks and protects data integrity and authenticity. Furthermore, the message counter and sequences with SHA-512 are integrated as part of the improved security method for CAN XL. Although the default security measures of CAN XL are still in the process of evolving, our method guarantees the freshness of messages by utilizing a cross-correlation function to maintain them within a specified time frame, thereby preventing replay attacks. This necessitated modifications to FPGA based architecture, which was validated using MATLAB and VHDL. This demonstrated its superiority over CAN FD's security measures in effectively mitigating replay and denial-of-service attacks. As contributions of this work introduces a non-linear sequence function to increase the unpredictability of freshness values, thereby preventing adversaries from predicting communication sequences or deducing message patterns. This method eliminates the necessity of transmitting freshness messages at predetermined intervals and enables multiple nodes to maintain unique freshness values, thereby further reducing vulnerabilities. To further reduce resource utilization, a lightweight XOR-based authentication tag was designed and integrated. This 1-byte tag combines the FV, one ciphertext byte, and a keyschedule byte to provide an integrity indicator with minimal hardware overhead. Despite its small size, the tag maintained high randomness (entropy ≈ 7.26 bits) and demonstrated 100% replay resistance during simulation, confirming that the underlying FV entropy ensures message integrity and uniqueness even under constrained hardware conditions. Our FPGA-in-the-loop (FIL) validation guarantees the real-time testing of cryptographic protocols in a hardware-accelerated environment, thereby facilitating the practical deployment of the security architecture for real-time embedded systems. In order to verify the protocol's robustness, we implement replay attacks that illustrate effective threat mitigation and protection.
A Study of Electrical Equivalent Circuit Battery Parameter and SOC Estimation Methods, and Optimum Charging Strategies for Safety and Longevity
(2026-01-01) Jarid, Saad; Das, Manohar; Toulabi, Mohammad; Wang, Xia; Wang, Zhe; El-Syed, Mohammed A
This dissertation investigates enhancements to battery charging techniques by addressing issues associated with Lithium-ion batteries, particularly those related to fast and safe charging. The research covers a series of interconnected topics that contribute to advancements in battery technology. One goal of this study is to improve the accuracy of battery modeling and parameter estimation, the other being development of new charging techniques, ultimately improving the reliability and longevity of Li-ion batteries.One of the primary objectives focuses on investigating the estimation of time-varying parameters of the Electrical-Equivalent Circuit (EEC) model, which is essential for accurate modeling of battery’s charging behavior. Two techniques were used for estimation: a direct continuous-time system identification technique based on State Variable Filtering (SVF) and the Indirect Discrete Time (IDT) technique. The results from both methods are compared to determine the most effective approach for accurate modeling. The second objective focuses on assessing the robustness of the parameter estimation techniques by integrating a dynamic, time-varying EEC model to enhance the accuracy of State-of-Charge estimation. Kalman filter in its various forms is utilized for SOC estimation. The third objective of this research is to create an improved Multi-stage Charging Current (MCC) profile. The goal is to optimize the charging current at each stage to reduce the total charge time while ensuring the safety of the battery by carefully monitoring its temperature during charging. To ensure efficient and effective charging of this technique without compromising the battery health, careful control of the charging current at each stage is required. The fourth objective of this study is to develop a continuous, exponentially decaying charging current profile. This technique addresses one of the shortcomings of MCC, by minimizing the stress fluctuations caused by abrupt shifts in current during transitions between stages. This optimized current profile offers a balance between fast charging and protecting the battery from potential damage.
Personalized Autonomous Vehicle Motion Control Using IRL And MPC
(2026-01-01) Zhou, Zhaodong; Chen, Jun; Radovnikovich, Micho; Li, Jia; Jones, Giselle S
Autonomous vehicle (AV) motion control has been widely studied to improve path tracking accuracy, driving comfort, and safety. Model predictive control (MPC) is a promising method for AV control because it can explicitly handle vehicle dynamics, constraints, and multi-objective optimization. However, conventional time-triggered MPC requires solving an optimal control problem at every time step, which creates a high computational burden for real-time implementation. In addition, traditional MPC usually depends on manually tuned cost weights and predefined reference trajectories, making it difficult to adapt the controller to individual driving preferences. This dissertation addresses these challenges by developing learning-based personalized and computationally efficient motion control methods for AVs. First, event-triggered MPC is investigated for AV path tracking, where the optimal control problem is solved only when a triggering condition is satisfied. A switching prediction model is further used to support MPC operation across different speed ranges. To improve the controller behavior during non-triggered intervals, a linear based inter-event feedback method is developed to update the steering command using the current vehicle state without solving a new optimization problem at every sampling step. Second, inverse reinforcement learning is used to learn personalized lane change behavior from expert demonstrations. Interpretable trajectory features are designed to represent driving comfort, efficiency, lateral position, heading behavior, yaw response, and steering effort. The learned cost weights are used for personalized lane change path generation and are further incorporated into MPC to generate personalized lane change maneuvers without manual cost tuning. The proposed methods are validated using CARLA simulation, real-vehicle path-tracking experiments, scale-vehicle experiments, and truck on-road lane change testing. Results show that the proposed framework can reduce computational demand while maintaining effective tracking performance, improve inter-event feedback behavior, and generate lane change trajectories that closely match individual driver behavior.
Hybrid 5G–Satellite Adaptive Video Streaming for Low-Latency Cloud-Based Cooperative Perception
(2026-01-01) Alkharabsheh, Ekhlass Mesleh; Alawneh, Shadi; Rawashdeh, Osamah; Wardat, Mohammad; Li, Li
The growing complexity of autonomous and connected vehicle systems necessitates reliable, low-latency communication between vehicles, infrastructure, and cloud resources. This dissertation presents a comprehensive framework for optimizing real-time data transmission in cloud-based cooperative perception systems (CPS), focusing on adaptive video compression, hybrid networking, and cloud-based inference. The research aims to reduce end-to-end latency while maintaining high perception accuracy and connectivity reliability across heterogeneous communication environments. The proposed architecture integrates adaptive H.265 (HEVC) compression, hybrid 5G–Starlink networking, and GPU-accelerated inference to support scalable roadside unit (RSU) camera deployments. An adaptive compression controller dynamically adjusts frame rate, resolution, and bitrate in response to bandwidth fluctuations, while a hybrid link manager employs make-before-break switching and multipath transmission to sustain continuous connectivity. The system was evaluated through a combination of simulation, laboratory bench testing, and real-world field experiments under diverse network conditions and video workloads. Results demonstrate that the integrated approach achieves end-to-end latency below 100 milliseconds and connectivity uptime exceeding 99 percent, outperforming conventional single-network and fixed-parameter configurations. Statistical analyses confirm significant reductions in latency variance, frame loss, and recovery time following network disruptions. Moreover, the proposed methods maintain detection accuracy (mAP > 0.75) and system stability under high-load scenarios involving up to 100 concurrent RSU streams. This work contributes (i) a validated, modular framework for real-time cloud-assisted cooperative perception, (ii) empirical benchmarks quantifying compression–network trade-offs, and (iii) deployment guidelines for scalable CPS implementations in intelligent transportation systems. Collectively, these contributions advance the state of knowledge toward deployable, cloud-integrated Level-5 autonomous driving infrastructures, enabling robust perception pipelines with demonstrable reliability, scalability, and cost efficiency.
Resilience Enhancement of Post-Disaster Power Distribution Systems Using Deep Reinforcement Learning
(2026-01-01) Alotaibi, Raed Awadh M; Zohdy, Mohamed; Kaur, Amanpreet; Alghamdi, Ali; Edwards, William; Al-Salman, Zeina
Weather-driven extreme events are placing growing stress on aging distribution infrastructure and increasingly threaten continuity of service for critical loads during prolonged outages. Microgrids can enhance resilience by transitioning to islanded operations and supplying prioritized loads with local distributed energy resources (DERs); however, post-disaster restoration remains challenging because operators must make coupled discrete–continuous decisions under tight resource and operating constraints.This dissertation addressed this challenge by developing a parameterized deep reinforcement learning controller, PDQN-CLR, that targets priority-weighted restoration while enforcing operational feasibility under scarcity. The PDQN-CLR modeled restoration as a hybrid action in which a discrete operational category was selected and paired with a continuous parameter vector specifying the DER real and reactive power setpoints. The approach was evaluated in a closed-loop OpenDSS environment using the IEEE 123-node feeder configured as an islanded microgrid with five grid-forming battery energy storage systems and 17 prioritized critical loads over a 72-step (36-hour) horizon with a 30-minute decision interval. Snapshot power-flow evaluation was performed at each step. Uncertainty was represented through capacity-factor derating under sufficient and scarce regimes, and each episode randomized the fault scenario, derating level, and initial state of charge. This dissertation also introduced (i) an uncertainty-aware evaluation protocol based on capacity-factor derating; (ii) a three-tier priority-weighted reward to encode the critical-load hierarchy; and (iii) a DER-aware load service rule that reduced voltage-only overstatement in islanded operation. With sufficient resources, PDQN-CLR achieved a mean PCS of 0.94 versus 0.80 for a Greedy baseline, while maintaining a low constraint violation score (CVS) of approximately 0.009 in both sufficient and scarce regimes. The baseline produced substantially larger violation magnitudes (CVS ≈ 0.388–0.667), indicating more severe and/or more frequent exceedances of the constraints. These results indicate that parameterized deep reinforcement learning can improve priority-weighted restoration when capacity is available and preserve feasibility as a primary outcome when scarcity limits achievable restoration.
Utilizing XR Technology to Deliver Information to Participants Through Visual, Auditory, and Haptic Sensory Channels
(2026-01-01) Athamnah, Solaf Mohammad Yousef; Louie, Wing Yue Geoffrey; Louie, Wing Yue Geoffrey; Rawashdeh, Osamah; Alawneh, Shadi
In complex, high-stakes tasks such as remote drone and vehicle operation, effective Human-Machine interfaces are critical for ensuring operators comprehend task-critical information and maintain high situation awareness. However, flooding operators with information can overwhelm their senses, causing critical details to be lost. This raises the question: how should designers determine which sensory channel best suits each information cue? A 3×3×3 study was conducted, with information-to-sensory-channel mapping as a between-subjects factor and task type as a within-subjects factor. Tasks were performed in an Unreal Engine simulation. Preliminary analysis of 18 participants revealed that the visual channel consistently outperformed auditory and haptic channels in both accuracy and reaction time. Additionally, the nature of the information cue had an effect on comprehension. These findings demonstrate that effective multimodal interface design requires strategically mapping information cues to appropriate sensory channels rather than distributing them arbitrarily.
Design and Optimization of Low-Cost Multi-Band mmWave Antennas for Emerging V2X Applications
(2026-01-01) Pervez, Mohammad Shahed; Kaur, Amanpreet; Aloi, Daniel N; Olawoyin, Richard; Tyberkevych, Vasyl
The rapid evolution of intelligent transportation systems has intensified the demand for reliable, high-data-rate, and low-latency vehicular communication technologies. Vehicle-to-Everything (V2X) communication, encompassing Dedicated Short-Range Communication (DSRC), fifth-generation (5G) millimeter-wave (mmWave), and emerging sixth-generation (6G) systems, requires compact, high-performance, and cost-effective antenna solutions capable of operating under stringent automotive constraints. This dissertation presents the design, optimization, and validation of low-cost multi-band mmWave antenna systems tailored for emerging V2X applications.The research focuses on the development of compact, multi-band, and multiple-input multiple-output (MIMO) antenna architectures operating across DSRC (5.9 GHz), 5G mmWave (28 GHz and 38 GHz), and prospective 6G sub-terahertz frequency bands. A series of novel antenna designs are proposed, including dual-band patch antennas, compact MIMO arrays, beamforming-enabled mmWave arrays, conformal flexible antennas, electromagnetic bandgap (EBG)-assisted isolation structures, defected ground structures (DGS), and dual-mode leaky-wave antennas for beam scanning applications. Emphasis is placed on achieving wide impedance bandwidth, high radiation efficiency, low envelope correlation coefficient (ECC), high diversity gain, and enhanced mutual coupling suppression within compact and low-profile form factors suitable for vehicular integration. Comprehensive electromagnetic modeling and optimization are performed using ANSYS HFSS, while MATLAB is employed for post-processing and MIMO performance evaluation. Where applicable, simulated results are validated through experimental measurements, demonstrating close agreement and confirming the robustness and practical feasibility of the proposed designs. The antenna systems exhibit wide −10 dB impedance bandwidths, stable radiation characteristics, high isolation levels exceeding −20 dB in MIMO configurations, and beam steering or scanning capabilities suitable for dynamic vehicular environments. In addition to performance optimization, this work addresses critical practical considerations, including fabrication complexity, material selection, scalability to massive MIMO systems, regulatory compliance, and long-term automotive deployment. The results demonstrate that low-cost, compact, and scalable antenna architectures can effectively support current DSRC and 5G V2X requirements while providing a viable technological pathway toward future 6G and joint communication–sensing platforms. The contributions of this dissertation advance the state of the art in vehicular antenna engineering and provide a comprehensive framework for the design of next-generation automotive V2X antenna systems, enabling reliable connectivity, enhanced sensing capabilities, and intelligent transportation infrastructures.
Development of an AI-driven Predictive Model for Vehicle Feature Optimization & Diagnostic Tuning Using an Extended Kalman Filter (EKF) State Estimator
(2026-01-01) Mezaael, Abraham; Zohdy, Mohamed; Louis, Steven; Kaur, Amanpreet; Pappas, George
Modern connected and automated vehicles rely on increasingly complex sensing, perception, and estimation architectures to infer internal system states and predict future behavior from high-volume, noisy, and nonlinear sensor data. Similar challenges arise across a broad class of engineered systems and intelligent devices, including robotic and autonomous platforms, where accurate perception and prediction under uncertainty are essential for safe, reliable, and sustained operation. As these systems age or operate across diverse environments and usage conditions, discrepancies emerge between modeled dynamics and observed measurements due to sensor drift, component degradation, environmental variability, and unmodeled nonlinear effects. These discrepancies degrade state estimation accuracy, destabilize diagnostic interpretations, and reduce the reliability of perception-driven and prediction-driven decision-making. This dissertation develops an adaptive perception and prediction framework that integrates nonlinear state estimation with data-driven learning to support continuous v system understanding under uncertainty. The proposed approach combines a model-based estimation layer that captures nominal system dynamics and enforces physical consistency with a learning-based predictive layer that models residual nonlinearities, long-term signal drift, and cross-sensor dependencies that are difficult to represent analytically. Through recursive estimation and forward prediction, the framework enables simultaneous inference of latent system states and anticipation of future system behavior. An adaptive tuning mechanism further refines perception and diagnostic interpretations by dynamically adjusting internal thresholds and confidence bounds based on predicted behavior, operational context, historical system performance, and evolving system characteristics. The framework is evaluated using both real-world and simulated datasets representative of connected vehicle networks and autonomous sensing systems. The evaluation demonstrates accurate estimation and prediction of motion states, actuation responses, thermal behavior, and system health indicators under varying operating conditions. Experimental results show improved estimation accuracy, enhanced predictive stability, faster convergence, and increased robustness to noise and sensor degradation when compared to conventional model-based estimators, standalone learning-based approaches, and static diagnostic frameworks. The proposed methodology further enables more consistent perception and diagnostic interpretation across heterogeneous systems, devices, and operating environments. This research contributes a scalable and generalizable methodology for integrated perception and prediction in modern connected vehicles and autonomous systems. By vi enabling estimation and diagnostic models to continuously learn, adapt, and recalibrate over time, the proposed framework supports long-term autonomy, resilience, and reliability in complex cyber-physical systems operating under uncertainty, change, and degradation. The proposed framework supports next-generation predictive maintenance strategies in connected fleets and contributes to the realization of adaptive, self-calibrating diagnostic systems for intelligent vehicles.
Learning more from limited demonstrations: methods for efficient and informative human-robot interaction
(2025-01-01) Chen, Qinghua; Rawashdeh, Osamah A.; Rawashdeh, Osamah A. A.; Louie, Wing-Yue Geoffrey; Korneder, Jessica; Qu, Guangzhi; Wang, Yanfeng
Learning from Demonstration (LfD) offers a promising paradigm for enabling socially assistive robots (SARs) to acquire complex skills and social behaviors by observing human demonstrations. However, conventional LfD methods rely heavily on large-scale, high-quality demonstration datasets, which are difficult to obtain in realworld healthcare and education settings due to privacy, cost, and data scarcity. This dissertation addresses these challenges by proposing a series of approaches to enhance learning efficiency, improve adaptability, and maximize information extraction from limited demonstration data in human-robot interaction (HRI) scenarios.First, a hierarchical deep reinforcement learning framework is introduced that incorporates auxiliary classifier generative adversarial networks (ACGAN), dynamic experience replay strategies, and Deep Q-learning Networks (DQN) to improve learning performance without additional data. Second, a task-oriented Meta-inverse reinforcement learning (Meta-IRL) approach is proposed to enhance adaptation to new tasks by leveraging encoders and exploring transformer-based multi-head and layer feature extraction strategies. Finally, a novel framework integrating a Global Attention Mechanism (GAM) with multi-layer feature fusion and latent Dirichlet allocation (LDA) topic modeling is developed to enrich feature representations and optimize prompt generation in few-shot learning. Experimental validation on robot-mediated therapy tasks and other datasets demonstrates that the proposed methods enhance performance under limited data conditions. Overall, this work integrates efficient learning mechanisms with personalized intervention strategies, enabling the model to acquire richer and more informative representations that enhance its performance, while contributing to the broader goal of facilitating effective deployment of intelligent SARs in data-constrained, real-world environments.
Adaptive deep canonical correlation analysis–based multimodal sentiment analysis
(2025-01-01) Liao, Yunhong; Li, Jia; Louie, Wing-Yue Geoffrey; Qu, Hongwei
Emotion recognition plays a critical role in affective computing and human-computer interaction. While physiological signals such as electroencephalography (EEG) and eye-tracking offer valuable insights into emotional states, effectively fusing these heterogeneous modalities remains challenging due to differences in temporal scale, dimensionality, and signal characteristics. Traditional fusion methods employ fixed strategies that fail to adapt to dynamic changes in modality reliability and cross-subject variability, limiting their practical applicability.
Evaluation of human-machine interfaces with varying haptic fidelity levels in AR/VR applications
(2025-01-01) Al-Shubeilat, Fares Tareq; Rawashdeh, Osamah; Louie, Wing-Yue Geoffrey; Alawneh, Shadi
This thesis evaluates how varying haptic fidelity levels shape the user experience and performance in gesture-based touchscreen interactions conducted in immersive VR. A between-subjects design compared the four conditions: a benchmark of real-world tablet touchscreen and three VR configurations, no haptic fidelity (hand tracking only), low haptic fidelity (vibrotactile gloves), and high haptic fidelity (tactile + force feedback). The participants completed touchscreen tasks (tap, swipe, pan, pinch) and a combinational task while subjective outcomes (presence, embodiment, and system usability) and objective performance metrics were recorded. The study answers three questions: (i) whether haptic fidelity alters perceived experience, (ii) how fidelity influences task performance, and (iii) which gestures are most sensitive to the haptic fidelity. The results show that perceived metrics were indistinguishable across the three haptic fidelity conditions, suggesting high quality consistent visuals/interactions dominated over the incremental in the touch cues richness at the fidelities tested.
Augmented reality for multimodal ultrasound-based breast biopsy
(2025-01-01) Hassan, Yasmeen; Wiacek, Alycen; Mirza, Khalid; Li, Jia
Augmented Reality (AR) technologies have been demonstrated to enhance image-guided medical procedures by improving real-time visualization and spatial understanding. Ultrasound, in particular, is often integrated with AR due to its real-time functionality and portability. However, most ultrasound-based AR systems are limited to traditional B-mode ultrasound alone, visualizing the tissue morphology, but lacking mechanical or microsctructural properties of tissue. Newer modalities such as elastography and Quantitative Ultrasound (QUS) can provide these additional properties to improve diagnosis. This thesis presents an AR platform that integrates multimodal ultrasound imaging, including elastography and QUS, into a spatially registered system for biopsy guidance. The system combines a Clarius handheld ultrasound probe with a Microsoft HoloLens 2 using Unity, OpenCV, and marker-based tracking. A Qt and MATLAB-based data pipeline supports streaming and spatial alignment of multiple imaging modes. System evaluation on tissue-mimicking phantoms showed registration errors of 2.8mm for B-mode at shallow depths, 10.8mm for B-mode in deeper tissues, and 17.0mm for elastography. Latency ranged from 159ms using B-mode to 167s for QUS imaging. These results demonstrate the feasibility of the proposed AR platform for ultrasound-guided interventions driven by both morphology and diagnostic information, while highlighting areas for future improvement in latency and registration accuracy and provides a foundation for future innovations in AR-based multimodal breast biopsies.
Advanced deep learning to generate and detect fake images of Egyptian monuments
(2025-01-01) Alaswad, Daniyah; Zohdy, Mohamed; Ganesan, Subramaniam; Louis, Steven; Solomonson, Bill
This study examined the use of StyleGAN to create synthetic images of Egyptian monuments, addressing a critical gap at the intersection of generative artificial intelligence and cultural heritage. Through extensive experiments on a large datasetcontaining 5,000 Egyptian monument images, we show that architectural changes to the StyleGAN framework can significantly improve the quality and authenticity of the generated images. Our study contributes to the existing literature. First, we designed an enhanced discriminator architecture incorporating noise injection, squeeze-and-excitation blocks, and an improved MinibatchStdLayer, resulting in a Fréchet Inception Distance 27.5 better than that of the original model. We further introduced a novel image-text alignment approach using SigLIP, which can generate semantically guided monuments. We applied Differential Evolution (DE) to optimize the latent space of the conditional generator to reduce the alignment error by 15 for the targeted monument-generation tasks. We systematically analyzed various truncation methods used to manage noise in generated images by finding the best parameters that fit the architecture best but are also diverse. Statistical validation using bootstrap confidence intervals, McNemar’s test and DeLong’s ROC analysis show significant improvements with effect sizes in the moderate to large range (Cohen’s d ≈ 0.9-1.4) The discriminator was able to achieve 95.5 accuracy with a 5.3 false positive rate and 3.6 false negative rate. This 62 error drop was compared to the baseline. Under heavily corrupted conditions (JPEG quality = 10; Gaussian blur σ = 5.0), it achieved 78-85 of the baseline performance, whereas the default achieved 65-72 of the baseline performance. Frequency domain analysis results revealed resilience, with AUC values generally >0.95, varying by frequency. The new discriminator was approximately 20 to 25 percent more robust to adversarial attacks. However, both architectures are fundamentally vulnerable to stronger attacks. Our research shows how strategic refinements of operations models can produce representations of Egyptian monuments that attain a high-quality and satisfactory level of diversity that we can detect. Innovations can greatly help in the preservation of cultural heritage, virtual tourism, visualization, and education. This study will allow the generation of high-quality and varied Egyptian monument images, which can help in the digital conservation and easy accessibility of one of the world’s great architectural heritages.
Self-driving vehicles as high-performance computing systems – a hybrid end-device to cloud approach
(2025-01-01) ABDELHAFIZ, AHMAD FAYEZ KHALED; Ganesan, Subramaniam; Alawneh, Shadi; Nezamoddini, Nasim; Schmidt, Darrell
The rapid evolution of autonomous-vehicle (AV) technology has defined a new era of intelligent transportation systems, promising improved safety, efficiency, and scalability. However, the exponential increase in the number of onboard sensors and the corresponding growth in machine-generated data have created challenges in real-time processing, cloud dependency, and sustainable data management. Traditional centralized cloud infrastructures struggle to process this volume efficiently, motivating the need for distributed, edge–cloud hybrid computing paradigms.This research introduces a hybrid edge-to-cloud framework in which autonomous vehicles utilize their onboard supercomputers when underused or idle as distributed processing nodes. These vehicles collectively perform computational tasks such as lane feature extraction, harvesting data, HD-map generation, and update, operating as a parallel network of edge machines that seamlessly integrates with the cloud for aggregation and scalability. The proposed system demonstrates this concept through a crowdsourced HD-map creation pipeline that fuses lane level data from multiple vehicles and integrates it with open geographic information from OpenStreetMap (OSM). The system architecture includes edge-based preprocessing, cloud-based multi-vehicle fusion, and map redistribution. Each vehicle locally extracts lane features from camera and GPS data and transmits compact representations to the cloud. The cloud aligns and fuses these extractions using geometric registration, probabilistic averaging, and version-controlled map management. Experimental results on a self-collected three lane highway dataset achieved absolute accuracy of 0.98–1.15m RMSE (2σ ≤ 1.35m) and relative inter-lane accuracy of 0.11–0.13m mean |ΔW| (2σ ≤ 0.18m), satisfying commercial HD-map standards. Beyond HD mapping, this study illustrates the broader potential of autonomous vehicle data harvesting, where the collective computational and sensing power of a fleet can be leveraged for large scale tasks. The findings demonstrate that a camera-only, hybrid edge–cloud approach can achieve LiDAR comparable mapping precision on highway while significantly reducing the cost and cloud dependency, paving the way toward higher levels of autonomy within smarter and more sustainable transportation networks.
Blockchain-integrated smart grid architecture with FBG sensing and EKF for fault and anomaly detection
(2025-01-01) Al Dakhl, Surah Adel; Zohdy, Mohamed; Li, Jia; Monroe, Ryan; Schmidt, Darrell
The smart grid has emerged to address the shortcomings of one-way existing grid systems and is the next generation power grid infrastructure that applies smart ICT (Information Communication Technology) to existing grid. The Smart Grid is expected to greatly improve the efficiency and reliability of future power systems with the demand for renewable energy resources. However, because major power facilities are interconnected through communication networks, Smart Grids cyber security is becoming an important issue. Cyber-attacks by malicious intruders can lead to serious incidents such as massive outages and the destruction of power network infrastructure, since cyber-attacks can damage energy data, starting with personal information leakage from grid members. Therefore, as a solution to this issue we will suggest a secure smart energy management system based on the blockchain. The combination of blockchain technology, optical fiber sensors, and Kalman filters in smart grids holds great potential for the future of power transmission and distribution systems. Blockchain offers secure protection against cyber threats and unauthorized access, while optical fiber sensors provide real-time monitoring and control of electrical energy flow. The integration of these technologies leads to improved transparency in energy generation, distribution, and consumption. Extended Kalman filters are utilized to identify and minimize uncertainties in data collected from optical fiber sensors, thereby enhancing the accuracy of information used for energy management and grid control. This integration promises increased security, enhanced reliability, improved efficiency, and greater flexibility in energy management. This paper presents a comprehensive examination of the benefits and limitations of integrating blockchain technology, optical fiber sensors, and Kalman filters in smart grids.
Interface and processing circuit design for ionic liquid hydrogen sensors
(2025-01-01) LING, SHUAISHUAI; Qu, Hongwei; Li, Jia; Kaur, Amanpreet
Hydrogen detection is essential for fuel-cell safety, hydrogen storage, and leak prevention. Ionic-liquid electrochemical sensors generate ultra-low currents (pA--nA), requiring highly sensitive acquisition. This thesis work developed a precision analog front end featuring a potentiostat for stable electrode bias and a transimpedance amplifier with selectable feedback up to 1GOhm for reliable current-to-voltage conversion. Input biasing and low-pass filtering improve stability and noise suppression. Although designed for IL-based H2 sensors, the system is adaptable to other platforms requiring pico- to nano-amp current measurement.
Automated parking systems using reinforcement learning-assisted model predictive control
(2025-01-01) Alawsi, Hussein Ali; Chen, Jun; Cheok, KaC; Radovnikovich, Micho; Schmidt, Darrell
Autonomous parking remains a challenging task due to the need for accurate trajectory tracking, smooth steering, and stable heading control under diverse maneuvering conditions. Conventional model predictive control (MPC) can handle system constraints effectively, but its performance depends heavily on manually tuned cost weights. This dissertation proposes a reinforcement learning-assisted model predictive control (RL-assisted MPC) framework to improve autonomous vehicle parking performance. A Deep Q-Network (DQN) agent is trained to dynamically select the cost function weights of an MPCcontroller, enabling real-time adaptation based on the vehicle’s current state. The hybrid approach leverages the predictive optimization capability of MPC together with the adaptive decision-making of RL, enabling the controller to adjust trade-offs in real time without manual re-tuning. The framework is evaluated across five different parking scenarios and compared against static-weight MPC baselines. Experimental evaluations demonstrate that the proposed RL-assisted MPC framework achieves comparable or better lateral tracking accuracy, while consistently providing smoother steering behavior and improved heading stability compared to baseline controllers using static MPC weights. The results demonstrate that RL-assisted MPC improves robustness and generalization in automated parking systems, highlighting the potential of combining model-based predictive control with RL for autonomous driving.
Towards AI-driven socially assistive virtual robots for personalized early childhood education: an investigation of feasibility, usability, and parent-supervised configuration
(2025-01-01) Abbas, Ibrahim; Ganesan, Subramaniam; Rawashdeh, Osamah; Debnath, Debatosh; Qu, Harvey
This paper presents an extensive investigation into the adoption of AI-driven Socially Assistive Virtual Robots (SAVRs) in simulation-based environments for early childhood education. By integrating child-focused feasibility data (engagement, speech recognition, success rates) with parent-focused usability data (task configuration challenges, speech misinterpretations, AI clarity), we analyze how eight children (ages 3–4) and ten parents used a ChatGPT-powered simulation framework to practice counting and letter recognition at home.Child results show that 4-year-olds achieved near-perfect task completion with minimal frustration, while 3-year-olds faced more difficulties due to incomplete articulation, shorter attention spans, and repeated speech recognition errors. Parent findings reveal moderate setup complexity (40 described it as “tedious”), frequent manual overrides for speech errors (70 cited speech recognition as a major frustration), and a strong preference (80) for cost-effective simulation over expensive physical SARs, provided usability improvements are made. We explore literature on Socially Assistive Robots (SARs), child-specific speech recognition challenges, AI-based adaptive learning, and prior human-robot interaction (HRI) studies, situating our work in the context of both physically embodied SARs and simulation-based agents that increasingly fulfill similar pedagogical roles [1], [2], [3], [4]. The methodology details the CoppeliaSim environment, ChatGPT-based real-time adaptation, multi-threaded speech orchestration, and the parent configuration interface. The results section includes tables (I–III) summarizing child performance, supplemented with in-depth analysis. We then present a discussion linking child engagement patterns to parent usability concerns, culminating in recommendations for child-specific ASR, multimodal input, short session durations, wizard-style interfaces, and context-aware AI prompts. Future directions explore hybrid physical-virtual usage, specialized acoustic modeling for 3-year-olds, and scaling to larger, more diverse samples. Overall, the study addresses child feasibility and parent usability in a unified manner, underscoring how speech recognition and user-friendly design can support wide-scale implementation of AI-driven SAVRs at home.

Browse

Recent Submissions