preprints

1. TechRxiv

   Towards Trustworthy AI: Real-Time Uncertainty Monitoring and Adversarial Detection

   Chern Chao Tai, Abhijeet Solanki, Wesam Al Amiri, Douglas A. Talbert, Syed Rafay Hasan, and Terry N. Guo

   TechRxiv 2025

   Abs Bib

   This paper presents a comprehensive approach to enhancing the trustworthiness of AI systems through real -time uncertainty monitoring and adversarial detection. We propose a novel framework that integrates Monte Carlo dropout and noise injection techniques to quantify model uncertainty, allowing for the identification of potential adversarial inputs. Our method is designed to operate in real-time, making it suitable for deployment in safety-critical applications such as autonomous vehicles and healthcare systems. We evaluate our approach using a variety of datasets and demonstrate its effectiveness in improving the robustness and reliability of AI models against adversarial attacks. The results highlight the importance of incorporating uncertainty analysis into AI systems to enhance their trustworthiness and ensure their safe deployment in real-world scenarios.

   @article{tai2025towards,
     title = {Towards Trustworthy AI: Real-Time Uncertainty Monitoring and Adversarial Detection},
     author = {Tai, Chern Chao and Solanki, Abhijeet and Al Amiri, Wesam and Talbert, Douglas A. and Hasan, Syed Rafay and Guo, Terry N.},
     journal = {TechRxiv},
     year = {2025},
     publisher = {TechRxiv},
   }

2. TechRxiv

   Input-Slice Attacks and Defense against Pseudo-Model-Parallelism-Based Collaborative DNN

   Abhijeet Solanki, Parth Patel, Faiq Khalid, Syed Rafay Hasan, and Uvais Qidwai

   TechRxiv Dec 2025

   Abs Bib

   This paper investigates input-slice attacks and defense mechanisms against pseudo-model-parallelism -based collaborative deep neural networks (DNNs). We analyze the vulnerabilities of collaborative DNNs to input-slice attacks, where adversaries manipulate specific slices of input data to deceive the model. We propose a novel defense strategy that incorporates robust training techniques and anomaly detection to mitigate the impact of such attacks. Our evaluation demonstrates the effectiveness of our defense mechanism in enhancing the resilience of collaborative DNNs against input-slice attacks, ensuring the integrity and reliability of AI systems in collaborative environments.

   @article{solanki2025input,
     title = {Input-Slice Attacks and Defense against Pseudo-Model-Parallelism-Based Collaborative DNN},
     author = {Solanki, Abhijeet and Patel, Parth and Khalid, Faiq and Hasan, Syed Rafay and Qidwai, Uvais},
     journal = {TechRxiv},
     year = {2025},
     month = dec,
     publisher = {TechRxiv},
   }

3. TechRxiv

   Vulnerability Assessment of Adversarial Image Attacks in a Miniaturized Autonomous Vehicle Testbed without Relying on Way-Point Information

   Tahmid Hasan Sakib, Abhijeet Solanki, Wesam Al Amiri, Syed Rafay Hasan, Syed Ali Asad Rizvi, and Terry N Guo

   TechRxiv Mar 2026

   Abs Bib

   This paper presents a vulnerability assessment of adversarial image attacks in a miniaturized autonomous vehicle testbed without relying on way-point information. We evaluate the susceptibility of the testbed to various adversarial image attacks, which can manipulate the perception system of autonomous vehicles. Our assessment focuses on the effectiveness of these attacks in deceiving the vehicle’s perception algorithms and the potential consequences on navigation and safety. The results highlight the importance of robust defense mechanisms to protect against adversarial image attacks in autonomous vehicle systems, particularly in scenarios where way-point information is not available for validation.

   @article{sakib2026vulnerability,
     title = {Vulnerability Assessment of Adversarial Image Attacks in a Miniaturized Autonomous Vehicle Testbed without Relying on Way-Point Information},
     author = {Sakib, Tahmid Hasan and Solanki, Abhijeet and Al Amiri, Wesam and Hasan, Syed Rafay and Rizvi, Syed Ali Asad and Guo, Terry N},
     journal = {TechRxiv},
     year = {2026},
     month = mar,
     publisher = {TechRxiv},
   }

4. TechRxiv

   Blinded by the Beam: A Unified Real-Time Defense Against Laser-Based Attacks on Navigational Perception of Autonomous Vehicles

   Abhijeet Solanki, Wesam Al Amiri, Syed Rafay Hasan, and Terry N Guo

   TechRxiv Sep 2025

   Abs Bib

   This paper presents a unified real-time defense mechanism against laser-based attacks on the navigational perception of autonomous vehicles. We propose a novel approach that combines hardware-level detection and software-level mitigation strategies to enhance the robustness of lidar-based perception systems. Our method is designed to operate in real-time, making it suitable for deployment in safety-critical applications. We evaluate our approach using a variety of datasets and demonstrate its effectiveness in improving the resilience of autonomous vehicle perception systems against laser-based attacks. The results highlight the importance of integrating multiple defense mechanisms to ensure the safety and reliability of autonomous vehicles in challenging environments.

   @article{solanki2025blinded,
     title = {Blinded by the Beam: A Unified Real-Time Defense Against Laser-Based Attacks on Navigational Perception of Autonomous Vehicles},
     author = {Solanki, Abhijeet and Al Amiri, Wesam and Hasan, Syed Rafay and Guo, Terry N},
     journal = {TechRxiv},
     year = {2025},
     month = sep,
     publisher = {TechRxiv},
   }

5. TechRxiv

   RAID: Real-Time XAI-Guided Adversarial Attacks for Edge-Deployed Neural Networks

   Abhijeet Solanki, Faiq Khalid, Drew Phelps, Syed Rafay Hasan, Terry Nan Guo, and Uvais Qidwai

   TechRxiv Feb 2026

   Abs Bib

   This paper introduces RAID, a novel framework for real-time explainable AI (X AI)-guided adversarial attacks on edge-deployed neural networks. RAID leverages XAI techniques to identify vulnerable regions in the input data and generate targeted adversarial examples that can deceive edge-deployed neural networks. We evaluate the effectiveness of RAID on various edge devices and demonstrate its ability to successfully bypass security measures while maintaining high visual fidelity. Additionally, we propose defense mechanisms to mitigate the impact of RAID attacks, including adversarial training and input preprocessing techniques. Our experimental results show that these defenses can significantly reduce the success rate of RAID attacks, enhancing the security of edge-deployed neural networks.

   @article{solanki2026raid,
     title = {RAID: Real-Time XAI-Guided Adversarial Attacks for Edge-Deployed Neural Networks},
     author = {Solanki, Abhijeet and Khalid, Faiq and Phelps, Drew and Hasan, Syed Rafay and Guo, Terry Nan and Qidwai, Uvais},
     journal = {TechRxiv},
     year = {2026},
     month = feb,
     publisher = {TechRxiv},
   }

peer reviewed

2026

1. ACCESS

   Blinded by the Beam: A Unified Real-Time Defense Against Laser-Based Attacks on Navigational Perception of Autonomous Vehicles

   Abhijeet Solanki, W. Al Amiri, S. R. Hasan, and T. N. Guo

   IEEE Access 2026

   Abs Bib

   This paper presents a unified, real-time defense against laser-based attacks that blind the camera and LiDAR perception of autonomous vehicles. We characterize how laser interference degrades navigational perception and introduce a lightweight detection-and-mitigation mechanism that neutralizes attacks before they reach the navigation stack, while remaining efficient enough to run directly on in-vehicle edge hardware such as the NVIDIA Jetson Orin and Nano. Experiments demonstrate reliable real-time detection with low latency, preserving perception integrity under adversarial physical-world conditions.

   @article{solanki2026blinded,
     title = {Blinded by the Beam: A Unified Real-Time Defense Against Laser-Based Attacks on Navigational Perception of Autonomous Vehicles},
     author = {Solanki, Abhijeet and Al Amiri, W. and Hasan, S. R. and Guo, T. N.},
     journal = {IEEE Access},
     volume = {14},
     pages = {57235--57255},
     year = {2026},
     publisher = {IEEE},
   }

2025

1. ACCESS

   Survey of Navigational Perception Sensors’ Security in Autonomous Vehicles

   Abhijeet Solanki, W. Al Amiri, M. Mahmoud, B. Swieder, S. R. Hasan, and T. N. Guo

   IEEE Access Jun 2025

   Abs Bib

   This paper presents a comprehensive survey of the security vulnerabilities and attack vectors associated with navigational perception sensors in autonomous vehicles (AVs). We analyze various types of sensors, including LiDAR, radar, cameras, and GPS, and discuss the potential threats they face from adversarial attacks. The survey categorizes attacks based on their target sensor and attack methodology, providing insights into the motivations behind these attacks and their potential impact on AV safety. We also review existing defense mechanisms and propose future research directions to enhance the security of navigational perception systems in AVs.

   @article{solanki2025survey,
     title = {Survey of Navigational Perception Sensors' Security in Autonomous Vehicles},
     author = {Solanki, Abhijeet and Al Amiri, W. and Mahmoud, M. and Swieder, B. and Hasan, S. R. and Guo, T. N.},
     journal = {IEEE Access},
     pages = {1--29},
     month = jun,
     year = {2025},
     publisher = {IEEE},
     doi = {10.1109/ACCESS.2025.3578891},
   }

2. SOUTHEASTCON

   ReAL: Machine Learning Detection of Reflective Attacks Against Lidarometry

   Abhijeet Solanki, L. Beirne, S. R. Hasan, and W. Al Amiri

   In SoutheastCon 2025 Mar 2025

   Abs Bib Code

   This paper introduces ReAL, a machine learning-based approach for detecting reflective attacks against lidarometry in autonomous vehicles. Reflective attacks involve the manipulation of lidar signals to create false perceptions of the environment, posing significant risks to AV safety. ReAL utilizes a combination of feature extraction and classification techniques to identify anomalies in lidar data that may indicate the presence of reflective attacks. We evaluate the performance of ReAL using a dataset of simulated and real-world lidar data, demonstrating its effectiveness in accurately detecting reflective attacks while minimizing false positives. The results suggest that ReAL can serve as a valuable tool for enhancing the security of lidar-based perception systems in autonomous vehicles.

   @inproceedings{solanki2025real,
     title = {ReAL: Machine Learning Detection of Reflective Attacks Against Lidarometry},
     author = {Solanki, Abhijeet and Beirne, L. and Hasan, S. R. and Al Amiri, W.},
     booktitle = {SoutheastCon 2025},
     pages = {1309--1313},
     month = mar,
     year = {2025},
     organization = {IEEE},
     doi = {10.1109/SoutheastCon56624.2025.10971487},
   }

3. SOUTHEASTCON

   GNAPing On the Job: Attacking and Defending Facial Detection on Edge Devices

   Abhijeet Solanki, R. T. Thornton, S. R. Hasan, and U. Qidwai

   In SoutheastCon 2025 Mar 2025

   Abs Bib Code

   This paper presents GNAPing, a novel approach for attacking and defending facial detection on edge devices. GNAPing exploits vulnerabilities in facial detection algorithms to create adversarial examples that can bypass detection systems. We evaluate the effectiveness of GNAPing on various edge devices, demonstrating its ability to successfully evade facial detection while maintaining high visual fidelity. Additionally, we propose defense mechanisms to mitigate the impact of GNAPing attacks, including adversarial training and input preprocessing techniques. Our experimental results show that these defenses can significantly reduce the success rate of GNAPing attacks, enhancing the security of facial detection systems on edge devices.

   @inproceedings{solanki2025gnaping,
     title = {GNAPing On the Job: Attacking and Defending Facial Detection on Edge Devices},
     author = {Solanki, Abhijeet and Thornton, R. T. and Hasan, S. R. and Qidwai, U.},
     booktitle = {SoutheastCon 2025},
     pages = {918--923},
     month = mar,
     year = {2025},
     organization = {IEEE},
     doi = {10.1109/SoutheastCon56624.2025.10971676},
   }

4. MWSCAS

   Realistic GPS Spoofing via Customized CARLA GPS Navigation and Controller Systems

   T. D. Robertson, W. Al Amiri, Abhijeet Solanki, S. R. Hasan, and T. Guo

   In 2025 IEEE 68th International Midwest Symposium on Circuits and Systems (MWSCAS) Aug 2025

   Abs Bib

   This paper presents a method for realistic GPS spoofing using customized CARLA GPS navigation and controller systems. GPS spoofing is a significant threat to autonomous vehicles, as it can lead to incorrect positioning and navigation. We utilize the CARLA simulator to create a controlled environment for testing GPS spoofing techniques and evaluate their effectiveness in deceiving autonomous vehicle navigation systems. Our results demonstrate the potential risks associated with GPS spoofing and emphasize the importance of developing robust countermeasures to protect against such attacks.

   @inproceedings{robertson2025realistic,
     title = {Realistic GPS Spoofing via Customized CARLA GPS Navigation and Controller Systems},
     author = {Robertson, T. D. and Al Amiri, W. and Solanki, Abhijeet and Hasan, S. R. and Guo, T.},
     booktitle = {2025 IEEE 68th International Midwest Symposium on Circuits and Systems (MWSCAS)},
     pages = {720--724},
     month = aug,
     year = {2025},
     organization = {IEEE},
     doi = {10.1109/MWSCAS53549.2025.11244479},
   }

5. ISCAS

   Mitigation of Camouflaged Adversarial Attacks in Autonomous Vehicles–A Case Study Using CARLA Simulator

   Yago Romano Martinez, Carter Brady, Abhijeet Solanki, Wesam Al Amiri, Syed Rafay Hasan, and Terry N. Guo

   In 2025 IEEE International Symposium on Circuits and Systems (ISCAS) Aug 2025

   Abs Bib

   This paper presents a case study on the mitigation of camouflaged adversarial attacks in autonomous vehicles (AVs) using the CARLA simulator. Camouflaged adversarial attacks involve the manipulation of visual inputs to deceive AV perception systems, particularly in traffic sign recognition (TSR). We evaluate the effectiveness of various mitigation strategies, including adversarial training and input preprocessing techniques, in enhancing the resilience of AV perception systems against such attacks. Our results demonstrate that these mitigation strategies can significantly reduce the success rate of camouflaged adversarial attacks, improving the security and safety of autonomous vehicles.

   @inproceedings{11044132,
     author = {Martinez, Yago Romano and Brady, Carter and Solanki, Abhijeet and Amiri, Wesam Al and Hasan, Syed Rafay and Guo, Terry N.},
     booktitle = {2025 IEEE International Symposium on Circuits and Systems (ISCAS)},
     title = {Mitigation of Camouflaged Adversarial Attacks in Autonomous Vehicles–A Case Study Using CARLA Simulator},
     year = {2025},
     volume = {},
     number = {},
     pages = {1-5},
     keywords = {Circuits and systems;Prevention and mitigation;Object detection;Cyber-physical systems;Cameras;Delays;Security;Artificial intelligence;Autonomous vehicles;Resilience;Autonomous vehicles (AVs);camera perception;camouflaged adversarial attacks;traffic sign recognition (TSR);and CARLA simulator},
     doi = {10.1109/ISCAS56072.2025.11044132},
   }

6. FLAIRS

   Towards trustworthy ai: Analyzing model uncertainty through monte carlo dropout and noise injection

   Chernchao Tai, W. Al Amiri, A. Solanki, D. A. Talbert, T. N. Guo, and S. R. Hasan

   In Proc. of The International FLAIRS Conference Aug 2025

   Abs Bib

   This paper explores the analysis of model uncertainty in the context of trustworthy AI through the use of Monte Carlo dropout and noise injection techniques. We investigate how these methods can be applied to enhance the reliability and robustness of AI models, particularly in safety-critical applications. Our study evaluates the effectiveness of these techniques in quantifying uncertainty and improving model performance under various conditions. The results demonstrate that incorporating uncertainty analysis can lead to more trustworthy AI systems, providing valuable insights for future research in this area.

   @inproceedings{tai2025towardsFLAIR,
     title = {Towards trustworthy ai: Analyzing model uncertainty through monte carlo dropout and noise injection},
     author = {Tai, Chernchao and Al Amiri, W. and Solanki, A. and Talbert, D. A. and Guo, T. N. and Hasan, S. R.},
     booktitle = {Proc. of The International FLAIRS Conference},
     year = {2025},
   }

7. Investigating Adversarial Image Attacks in a Sensor Fusion Framework Using a Scaled Autonomous Vehicle Testbed

   Tahmid Hasan Sakib, Wesam Al Amiri, Abhijeet Solanki, Syed Rafay Hasan, Terry N. Guo, and Syed Ali Asad Rizvi

   Aug 2025

2024

1. ISVLSI

   Investigate the Effects of Laser Attack on Intelligence of the AV Perception

   Abhijeet Solanki, S. R. Hasan, and T. N. Guo

   In 2024 IEEE Computer Society Annual Symposium on VLSI (ISVLSI) Aug 2024

   Abs Bib

   This paper investigates the effects of laser attacks on the intelligence of autonomous vehicle perception systems. Laser attacks can manipulate the data received by lidar sensors, leading to incorrect interpretations of the environment. We analyze the impact of such attacks on the performance of perception algorithms and propose mitigation strategies to enhance the robustness of these systems. The results demonstrate the vulnerability of current lidar-based perception systems to laser attacks and highlight the need for more resilient designs.

   @inproceedings{solanki2024investigate,
     title = {Investigate the Effects of Laser Attack on Intelligence of the AV Perception},
     author = {Solanki, Abhijeet and Hasan, S. R. and Guo, T. N.},
     booktitle = {2024 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)},
     pages = {779--782},
     year = {2024},
     organization = {IEEE},
     doi = {10.1109/ISVLSI61997.2024.00152},
   }

2022

2020

1. CBMS

   Protecting Electronic Health Records in Transit and at Rest

   Denis Ulybyshev, Christian Bare, Kristen Bellisario, Vadim Kholodilo, Bradley Northern, \textbfAbhijeet \textbfSolanki, and Timothy O’Donnell

   In 2020 IEEE 33rd International Symposium on Computer-Based Medical Systems (CBMS) Aug 2020

   Abs Bib 🏅Best Paper Award

   Electronic Health Records (EHRs) are a critical component of modern healthcare systems, containing sensitive patient information that must be protected from unauthorized access and data breaches. In this paper, we propose a comprehensive approach to securing EHRs both in transit and at rest. Our method combines cryptographic techniques, access control mechanisms, and data watermarking to ensure the confidentiality and integrity of EHRs. We evaluate our approach using a simulated healthcare environment and demonstrate its effectiveness in preventing unauthorized access and mitigating the risks associated with data breaches. Our results highlight the importance of robust security measures for protecting EHRs and ensuring compliance with regulations such as HIPAA.

   @inproceedings{9183319,
     author = {Ulybyshev, Denis and Bare, Christian and Bellisario, Kristen and Kholodilo, Vadim and Northern, Bradley and \textbf{Solanki}, \textbf{Abhijeet} and O'Donnell, Timothy},
     booktitle = {2020 IEEE 33rd International Symposium on Computer-Based Medical Systems (CBMS)},
     title = {Protecting Electronic Health Records in Transit and at Rest},
     year = {2020},
     volume = {},
     number = {},
     pages = {449-452},
     keywords = {Cryptography;Access control;Servers;Containers;Electronic medical records;Metadata;Watermarking;Electronic Health Records, data privacy, data leakage prevention, HIPAA, access control},
     doi = {10.1109/CBMS49503.2020.00091},
     award_name = {Best Paper Award},
   }

posters

1. SOUTHEASTCON

   Towards Machine Learning Based Fingerprinting of Ultrasonic Sensors

   Marim Elhanafy, Srivaths Ravva, Abhijeet Solanki, and Syed Rafay Hasan

   In SoutheastCon 2025 2025

   Abs Bib

   This paper presents a machine learning-based approach for fingerprinting ultrasonic sensors. Ultrasonic sensors are widely used in various applications, including robotics, automotive systems, and industrial automation. However, the security of these sensors is often overlooked, making them vulnerable to spoofing and other attacks. We propose a method that utilizes machine learning algorithms to analyze the unique characteristics of ultrasonic sensor signals and create a fingerprint for each sensor. Our approach involves feature extraction from the sensor signals, followed by classification using algorithms such as decision trees and random forests. We evaluate our method using a dataset of ultrasonic sensor signals collected from different sensors and demonstrate its effectiveness in accurately identifying and fingerprinting ultrasonic sensors. The results suggest that machine learning-based fingerprinting can enhance the security of ultrasonic sensor systems by enabling the detection of unauthorized or counterfeit sensors.

   @inproceedings{10971545,
     author = {Elhanafy, Marim and Ravva, Srivaths and Solanki, Abhijeet and Hasan, Syed Rafay},
     booktitle = {SoutheastCon 2025},
     title = {Towards Machine Learning Based Fingerprinting of Ultrasonic Sensors},
     year = {2025},
     volume = {},
     number = {},
     pages = {1332-1333},
     keywords = {Machine learning algorithms;Fingerprint recognition;Multilayer perceptrons;Acoustics;Manufacturing;Reliability;Object recognition;Decision trees;Character recognition;Random forests;sensor fingerprinting;sensor detection;machine learning-based sensor detection},
     doi = {10.1109/SoutheastCon56624.2025.10971545},
   }