Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Experimental Study of Trojan Vulnerabilities in UAV Autonomous Landing

Reza Ahmari, Ahmad Mohammadi, Vahid Hemmati, Mohammed Mynuddin, Mahmoud Nabil Mahmoud, Parham Kebria, Abdollah Homaifar, Mehrdad Saif

TL;DR

The paper investigates Trojan backdoor vulnerabilities in CNN-based autonomous UAV landing systems within Urban Air Mobility by conducting an empirical study on a DroNet-based framework using a self-collected landing-pad dataset with embedded triggers. It outlines a four-phase methodology (data collection, preparation, training, testing) and demonstrates a substantial performance degradation, with landing-zone accuracy dropping from $96.4\%$ to $73.3\%$ under Trojan activation. The results reveal concrete safety risks posed by backdoor attacks and provide a framework for evaluating Trojan vulnerability in real-time UAAV deployments, highlighting the need for lightweight detection and prevention strategies. Overall, the work contributes to securing UAM operations by quantifying Trojan risks and laying groundwork for resilient deep-learning-based navigation and landing systems.

Abstract

This study investigates the vulnerabilities of autonomous navigation and landing systems in Urban Air Mobility (UAM) vehicles. Specifically, it focuses on Trojan attacks that target deep learning models, such as Convolutional Neural Networks (CNNs). Trojan attacks work by embedding covert triggers within a model's training data. These triggers cause specific failures under certain conditions, while the model continues to perform normally in other situations. We assessed the vulnerability of Urban Autonomous Aerial Vehicles (UAAVs) using the DroNet framework. Our experiments showed a significant drop in accuracy, from 96.4% on clean data to 73.3% on data triggered by Trojan attacks. To conduct this study, we collected a custom dataset and trained models to simulate real-world conditions. We also developed an evaluation framework designed to identify Trojan-infected models. This work demonstrates the potential security risks posed by Trojan attacks and lays the groundwork for future research on enhancing the resilience of UAM systems.

An Experimental Study of Trojan Vulnerabilities in UAV Autonomous Landing

TL;DR

The paper investigates Trojan backdoor vulnerabilities in CNN-based autonomous UAV landing systems within Urban Air Mobility by conducting an empirical study on a DroNet-based framework using a self-collected landing-pad dataset with embedded triggers. It outlines a four-phase methodology (data collection, preparation, training, testing) and demonstrates a substantial performance degradation, with landing-zone accuracy dropping from to under Trojan activation. The results reveal concrete safety risks posed by backdoor attacks and provide a framework for evaluating Trojan vulnerability in real-time UAAV deployments, highlighting the need for lightweight detection and prevention strategies. Overall, the work contributes to securing UAM operations by quantifying Trojan risks and laying groundwork for resilient deep-learning-based navigation and landing systems.

Abstract

This study investigates the vulnerabilities of autonomous navigation and landing systems in Urban Air Mobility (UAM) vehicles. Specifically, it focuses on Trojan attacks that target deep learning models, such as Convolutional Neural Networks (CNNs). Trojan attacks work by embedding covert triggers within a model's training data. These triggers cause specific failures under certain conditions, while the model continues to perform normally in other situations. We assessed the vulnerability of Urban Autonomous Aerial Vehicles (UAAVs) using the DroNet framework. Our experiments showed a significant drop in accuracy, from 96.4% on clean data to 73.3% on data triggered by Trojan attacks. To conduct this study, we collected a custom dataset and trained models to simulate real-world conditions. We also developed an evaluation framework designed to identify Trojan-infected models. This work demonstrates the potential security risks posed by Trojan attacks and lays the groundwork for future research on enhancing the resilience of UAM systems.
Paper Structure (12 sections, 7 figures)

This paper contains 12 sections, 7 figures.

Table of Contents

  1. INTRODUCTION
  2. BACKGROUND & RELATED WORK
  3. TROJAN ATTACK CONCEPT AND IMPLEMENTATION
  4. TROJAN ATTACK IMPLEMENTATION
  5. PROPOSED METHODOLOGY
  6. OVERALL FRAMEWORK
  7. DATA COLLECTION
  8. DATA PREPARATION
  9. TRAINING PHASE
  10. TESTING PHASE
  11. RESULTS AND DISCUSSION
  12. CONCLUSIONS

Figures (7)

  • Figure 1: Trojan Attack Concept. Top: a small pattern on a road sign alters model prediction. Bottom: a trigger on a landing pad misguides landing zone detection, illustrating risks for UAM systems.
  • Figure 2: Overall framework for Trojan attack testing in the context of UAAVs. The diagram outlines the key phases: Data Collection, Data Preparation, Training, and Testing, illustrating how Trojan triggers are embedded into the dataset and tested on the model.
  • Figure 3: Custom dataset of varied landing pads simulating real-world urban UAV operating conditions.
  • Figure 4: Trojan triggers of 5x5, 10x10, and 15x15 chessboard patterns; 5x5 chosen for effective misclassification with minimal detectability.
  • Figure 5: Normal landing pads (top) vs. pads with embedded Trojan triggers (bottom) used to simulate attack scenarios.
  • ...and 2 more figures