Reinforcement Learning-Assisted Virtualized Security Framework for CAVs

Details

• Creators:
  Sahoo, Jagruti ; Mukwaya, Arthur Noa ; Ruganuza, Denis ; Benibo, Izison ; Gyimah, Nana Kankam ; Mwakalonge, Judith L. ; Comert, Gurcan ; Biswal, Biswajit ; Swain, Nikunja ; Fu, Yanzhou
• Corporate Creators:
  National Center for Transportation Cybersecurity and Resiliency (TraCR) National University Transportation Center (UTC) ; Clemson University
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
• Subject/TRT Terms:
  Autonomous Vehicles Buffer Overflow Code Diversification Computer Security Connected Vehicles Cybersecurity Learning Moving Target Indicators Network Function Virtualization (NFV) Networks Reinforcement Learning

  More +
• Publication/ Report Number:
  12
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report, 01/01/2024 - 05/31/2025
• Corporate Publisher:
  National Center for Transportation Cybersecurity and Resiliency (TraCR) National University Transportation Center (UTC)
• Abstract:
  Security vulnerabilities in the Connected and Autonomous Vehicle (CAV) software can allow hackers to perform malicious actions ranging from draining batteries and taking control of the steering wheel to disabling the alarm system. CAV software is subject to several cyber-attacks including memory corruption, code injection, remote code execution, and malware attacks. This study develops a Network Functions Virtualization (NFV)-based virtualized security framework to increase the resilience of CAV software. The proposed framework integrates a reinforcement learning-based code diversification mechanism that employs a Double Deep Q Network (DDQN) agent to optimally execute code variants of CAV software implemented as virtual network functions (VNFs). By meticulously switching the VNFs, the framework presents an unpredictable attack surface, thereby ensuring resilience. Our results show that the DDQN agent achieves a higher security access rate and higher uptime than a Q-learning agent, demonstrating the ability of the DDQN agent to provide an improved defense response and maintain continuity in the event of intrusions. Our prototype demonstrates the feasibility of hosting the CAV software programs as VNFs on a virtualized infrastructure. Moreover, we demonstrate the benefit of code diversification by switching the implementation of a point cloud concatenation functionality across different languages and simulating a memory corruption attack to test the program behavior. The research methods developed by this study can be applied to realize optimal software diversification in other areas such as the Internet of Things (IoT) and Cyber-physical Systems (CPS).
• Format:
  PDF
• Funding:
  69A3552344812, 69A3552348317
• Collection(s):
  FAA William J. Hughes Technical Center for Advanced Aerospace
• Main Document Checksum:
  urn:sha-512:4d5dae2160d78f45f2b927ae06a9a9523bbea498c02ad298b14ecb3d88330067fb740227ebb3c19d5300e7447b8c773b4e95bccab94fb5d034a6aa7f1633e28c
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/92506/dot_92506_DS1.pdf
• File Type:
  [PDF - 2.99 MB ]