Virginia Tech® home

Explainable Artificial Intelligence (XAI) Security for Logistic Disruption Mitigation in Distributed UAV Swarms and Other Autonomous Cyber-Physical Systems

Researchers aim to develop a machine learning (ML)-based security solution for distributed autonomous unmanned aerial vehicle (UAV) swarms and decentralized detection to provide strong mission assurance in contested environments.

Funded by the CCI Hub

Project Investigators
Rationale and Background

Cyber-physical systems (CPS) that integrate physical and digital spaces must be protected on critical infrastructures such as the Port of Virginia. 

For example, unmanned aerial vehicles (UAV) are used to monitor and communicate in remote reconnaissance missions, in surveillance operations, and to support command and control. Many applications are distributed, requiring coordination, planning, and runtime reconfiguration to conduct operations.

Traditionally, people manage UAV operations, but artificial intelligence (AI) has emerged as a critical tool. AI-based security controls that integrate ML algorithms into security domains (intrusion and malware detection) are considered more effective than traditional signature-based and heuristics-based controls. 

However, the growing adoption of advanced ML algorithms is turning these AI-based security controls into black-box systems.

Methodology

Researchers  will look at ML-based systems that require less human intervention and are more effective in detecting new attacks. 

They will focus on UAV swarm applications with decentralized and resource-limited nodes to address such areas as battery life challenges, developing detection mechanisms for adversarial actions in distributed UAV swarms. 

The proposed structure will enable them to consider complex scenarios, in which  UAVs might have inaccurate or faulty position information due to intrusion, noise, or measurement failure. 

Particular focus will be on transparency and interpretability and how they can be applied to an artificial neural network (ANN)-based CPS intrusion detection system (IDS).

Projected Outcomes

The main focus of this research is to provide techniques for more accountability in the use of algorithm-based controls for cybersecurity and provide directions for more explainable commands.  

Researchers aim to develop innovative data-driven approaches that employ a generative adversarial network (GAN), a class of deep machine learning,  that’s fully decentralized to overcome vulnerabilities to failure and intrusion. Objectives include: 

  • Simulating adversarial actions in a distributed UAV swarm environment.  
  • Developing resource-aware intrusion detection for UAV swarm, performance analysis, and risk mitigation. 
  • Building distributed ML applications and configurations. 
  • Building a platform with decentralized detection capability, predictive control, and failure prevention. 
  • Performing a system evaluation and a final demonstration.