PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey on Security Attacks and Defense Techniques for Connected and Autonomous Vehicles

Minh Pham, Kaiqi Xiong

arXiv:2007.08041v1cs.CR

TL;DR

CAVs combine autonomous driving with communications that expand their attack surface and can propagate impacts across vehicles and infrastructure. This survey synthesizes 184 papers and technical reports to classify attacks and defenses, examine attack models, and identify research challenges. It finds no strong connection between academic security research and industry implementation, with no evidence that marketed CAVs defend against some novel research attack models.

  • Problem

    CAV components are vulnerable to attacks, and connected vehicles can affect other CAVs and infrastructures through communication.

  • Method

    The paper surveys 184 papers from 2000 to 2020, organizing attacks and defenses by targeted components, access requirements, and attack motives.

  • Results

    The survey finds no strong connection between academic CAV security research and industry implementation, with no evidence that marketed CAVs defend against novel attack models identified by researchers.

  • Takeaways & Limitations

    The survey provides researchers and engineers with an organized account of CAV security issues, countermeasures, research gaps, and future directions.

Abstract

from arXiv · show

Autonomous Vehicle has been transforming intelligent transportation systems. As telecommunication technology improves, autonomous vehicles are getting connected to each other and to infrastructures, forming Connected and Autonomous Vehicles (CAVs). CAVs will help humans achieve safe, efficient, and autonomous transportation systems. However, CAVs will face significant security challenges because many of their components are vulnerable to attacks, and a successful attack on a CAV may have significant impacts on other CAVs and infrastructures due to their communications. In this paper, we conduct a survey on 184 papers from 2000 to 2020 to understand state-of-the-art CAV attacks and defense techniques. This survey first presents a comprehensive overview of security attacks and their corresponding countermeasures on CAVs. We then discuss the details of attack models based on the targeted CAV components of attacks, access requirements, and attack motives. Finally, we identify some current research challenges and trends from the perspectives of both academic research and industrial development. Based on our studies of academic literature and industrial publications, we have not found any strong connection between academic research and industry's implementation on CAV-related security issues. While efforts from CAV manufacturers to secure CAVs have been reported, there is no evidence to show that CAVs on the market have the ability to defend against some novel attack models that the research community has recently found. This survey may give researchers and engineers a better understanding of the current status and trend of CAV security for CAV future improvement.

I. INTRODUCTION

CAVs combine autonomous sensing with vehicle and infrastructure connectivity, improving transportation while expanding the opportunities and potential consequences of cyberattacks. This survey reviews the literature to organize attacks, defenses, research gaps, and development challenges.

  • CAV context: Automation levels 2 through 5 are considered because humans are not fully involved when automated driving features are engaged.
  • CAV context: CAVs integrate sensors and connection mechanisms that support navigation, vehicle communication, and infrastructure-assisted safety applications.Sensors include laser, radar, cameras, GPS, and LiDAR; connections include cellular, Bluetooth, WAVE, and Wi-Fi.
  • Security motivation: Connectivity and public deployment make CAVs attractive cyberattack targets, while limited driver availability can increase the potential harm of successful attacks.
  • Prior research: Prior research identified vulnerabilities in sensors, electronic control units, and connection mechanisms, including demonstrated attacks on components currently sold and operated.
  • Survey scope: 184 papers from 2000 to 2020 are surveyed to summarize attacks and defenses, organize attack models, and identify research gaps and challenges.The survey covers academic papers and technical reports and notes increasing publication activity from 2015 to 2019.

1) Attack targets:

The survey identifies attack targets across CAV interfaces, control systems, sensors, and communication networks. These components support essential driving functions but can also provide pathways for unauthorized data access, software modification, or malicious control.

  • OBD ports: OBD ports provide real-time vehicle data and pathways to access or modify ECU software, including during manufacturer firmware updates.
  • Electronic control units: ECUs control vehicle subsystems by processing component signals and sending control signals for functions such as braking, engine management, tire pressure, and inertial measurement.
  • CAN networks: CAN networks connect ECUs for high-speed communication, but malicious broadcasts can make the entire network vulnerable to cyberattacks.
  • Sensors: LiDAR measures surrounding-object distances with reflected light and supports obstacle detection and autonomous emergency braking.
  • Sensors: Radar supports functions including blind-spot monitoring, lane keeping, parking assistance, automatic distance control, and brake assistance.
  • Sensors: GPS supports autonomous route planning but lacks direct source authentication, leaving receivers vulnerable to jamming and spoofing.
  • Sensors: Cameras provide 360-degree vehicle views for traffic-sign recognition, lane detection, object detection, and distance measurement, but performance declines in rain, fog, or snow.

2) Classifications of Attack Models:

The paper classifies CAV attack models by access requirements and attacker motives, then organizes defenses into recurring strategy categories. These categories are useful guides, but defense effectiveness remains attack-specific.

  • Access Requirements: Attack access is classified as remote or physical, with CAN and ECUs vulnerable to both types.Remote attacks can send counterfeit data, block signals, or collect confidential data; physical attacks require modifying components or attaching instruments.
  • Attack Motivations: Attack motives are interrupting CAV operations, gaining control over vehicle movements, or stealing information.Examples include disabling autonomous driving, changing routes or speed, forcing emergency braking, and collecting confidential data.
  • Defense Categories: The survey groups defenses into anomaly-based intrusion detection, information redundancy, and encryption methods.Anomaly-based IDS detects abnormal signals or side-channel information, while redundancy combines sensors or external information and encryption protects unauthenticated CAN and sensor signals.
  • Defense Categories: Information redundancy can mitigate LiDAR jamming, radar jamming, and camera blinding but is ineffective against GPS jamming and increases cost.Multiple LiDAR sensors with different wavelengths can resist high-power light beams, although adding components is expensive.

III. EXISTING ATTACKS AND THEIR COUNTERMEASURES

The survey traces attack sequences across CAV components, showing how compromise of one component can propagate through ECUs, CAN, communication channels, and other vehicles. It therefore frames comprehensive component security as necessary.

  • Attack Sequences: A compromised CAV component can enable sequential attacks against other components, other CAVs, and infrastructure.The paper presents possible attack sequences connecting physical access, sensor compromise, telematics compromise, CAN messages, and compromised vehicles.
  • Attack Sequences: Physical access to an OBD port can lead to CAN access and subsequently to ECU compromise.This is the sequence labeled (1) in the paper’s attack-sequence taxonomy.
  • Attack Sequences: Compromised LiDAR, radar, GPS, or cameras can send adversarial information to ECUs.This is the sequence labeled (2).
  • Attack Sequences: Compromised telematics ECUs can transmit adversarial information through CAN to other ECUs.The relevant communication channels include VANET, Bluetooth, and DSRC.
  • Security Implication: Because attackers may compromise components one after another, manufacturers should secure all CAV components.The paper identifies attacks originating from an attacker’s CAV or from already compromised CAVs as another possible sequence.

A. Attacks on OBD

OBD ports expose CAVs to physical attacks that can reach CAN and ECUs, while existing defenses emphasize authentication, integrity, privacy, and downstream monitoring. The survey identifies unresolved challenges in securing the port itself and evaluating CAN defenses in vehicles.

  • OBD Attack Model: The OBD port is an open gateway because it commonly lacks encryption and access control, allowing physical attackers to reach other CAV components.OBD-connected devices may transfer data through wired or wireless connections, creating pathways into vehicle systems.
  • Defense Requirements: OBD defenses should authenticate devices, preserve their integrity and privacy, and complete authentication without significant user delay.The proposed criteria require trusted manufacturers, evidence that devices were not compromised, restricted information intelligibility, and efficient authentication.
  • Existing Defenses: The survey found no significant method for securing the OBD port or detecting malicious devices directly.Hardware-in-the-loop equipment supports attack and defense testing but does not itself provide an OBD defense layer.
  • CAN Attacks via OBD: CAN attacks through OBD include eavesdropping, replay or unauthorized transmission, and denial of service.These attacks rely on observing CAN messages, impersonating ECUs, or sending high-priority messages that prevent other CAN traffic from being processed.
  • CAN Countermeasures: Effective CAN defenses must provide confidentiality, authenticity, real-time performance, and low computing-resource requirements.The literature includes encryption, message-authentication approaches, and anomaly-based IDS methods using entropy, traffic patterns, or ECU-message timing.
  • Defense Challenges: CAN defenses remain difficult to select because proposed methods must balance real-time response, accuracy, vehicle modification, and legacy CAN constraints.The survey calls for comparative studies on moving vehicles to guide manufacturers.

C. Attacks on Electronic Control Units

ECU attacks may originate through CAN access or through connected telematics ECUs, while defenses include robust code, restricted access, and secure firmware updates. The survey notes that proposed FOTA frameworks still lack realistic validation.

  • ECU Attack Models: Attackers can compromise ECUs through CAN access obtained via an OBD port or telematics ECUs, then falsify inputs, inject code, or reprogram ECUs.The attack models distinguish direct CAN compromise from compromise through connection mechanisms such as Bluetooth or wireless access.
  • Defense Requirements: Defenses against telematics-ECU compromise should use robust firmware, trusted authenticated connections, and firmware updates with integrity and authenticity guarantees.The survey reports proposed FOTA protocols, while noting that installation can remain vulnerable to adversarial modification through TOCTTOU attacks.
  • Defense Challenges: No reviewed study implemented the proposed FOTA frameworks in a realistic CAV environment.The survey identifies validation of these theoretical frameworks as an outstanding research need.

D. Attacks on LiDAR

LiDAR attacks can spoof sensed object locations or jam legitimate returns, while defenses emphasize timely detection, signal filtering, and practical deployment constraints.

  • LiDAR spoofing: LiDAR spoofing can make real objects appear closer or farther by relaying or injecting counterfeit signals.A demonstrated implementation used two transceivers costing 49.9 US Dollars and triggered emergency braking.
  • LiDAR spoofing: Spoofing defenses include overlapping sensor views, randomized pulse directions or waveforms, and motion-based detection from prior LiDAR frames.Randomized waveforms can provide low cost and high immediacy, whereas random-direction transmission wastes pulses and reduces immediacy.
  • LiDAR jamming: LiDAR jamming overwhelms legitimate light with higher-intensity light at the same wavelength, preventing sensor acquisition.The survey reports a jamming demonstration but no experiment on a moving autonomous vehicle.
  • LiDAR jamming: Jamming countermeasures include V2V information, wavelength changes, multiple wavelengths, and shorter ping periods.The survey proposes low cost, high immediacy, and legitimate-signal filtering as defense criteria.
  • Challenges: Several LiDAR defenses remain insufficiently validated on CAVs, especially for moving vehicles and attacks using newer models.The survey specifically calls for testing proposed defenses against CAVs.

E. Attacks on radar

Radar attacks include signal spoofing and jamming, but CAV-specific feasibility and defense validation remain limited.

  • Radar spoofing: Radar spoofing uses replicated and rebroadcast signals to inject distorted sensor data, commonly through DRFM.Stored signal phase can be modified before rebroadcasting to the radar sensor.
  • Radar spoofing: Radar spoofing defenses seek timely attack detection, signal filtering, consistency, and non-disruptivity.PyCRA randomizes probing signals but can interrupt safety-critical components and produce inaccurate distances over time.
  • Radar spoofing: STCR transmits randomized-direction challenge signals and is reported to detect attacks while consistently measuring actual distance.Unlike PyCRA, it does not shut down sensing signals.
  • Challenges: Further experiments are needed because radar spoofing feasibility varies with sensor timing, and the leading defense had not been experimentally validated.The cited defense is described as state of the art but remains unvalidated experimentally.
  • Radar jamming: Radar jamming modifies stored signal frequency and amplitude to prevent detection of the object at the jammer’s location.The survey found no publication experimenting with this attack model on CAVs.

F. Attacks on GPS

GPS attacks spoof or overwhelm satellite signals; proposed defenses are generally simple, but their effectiveness against sophisticated attacks and CAV-specific conditions remains uncertain.

  • GPS spoofing: GPS spoofing broadcasts realistic counterfeit signals and gradually shifts their apparent location as receivers select the stronger signal.An attacker must calculate the victim distance with an error of at most 22.5 meters, though feasibility on moving CAVs remains unclear.
  • GPS spoofing: Seven basic countermeasures monitor signal strength, satellite identities, timing, and independent position checks.They are described as simple, inexpensive, interoperable, and requiring no satellite-transmitter modification.
  • GPS spoofing: These countermeasures may prevent simple attacks but can fall short against sophisticated GPS spoofing strategies.The survey identifies newer attack models as an unresolved defense challenge.
  • GPS jamming: GPS jamming overwhelms weak satellite signals, disrupting service as a denial-of-service attack.Unlike spoofing, jamming gives attackers less control over the receiver’s reported location.
  • GPS jamming: Anti-jamming approaches include adaptive-array processing, LMS filtering, and Turbo Coding to recover legitimate signals.The leading cited Turbo Coding method requires satellite modification and works only when jamming is not too strong.

G. Attacks on cameras

Camera attacks blind sensors or perturb images to mislead CAV vision models, while defenses emphasize low cost, generalization, and real-time operation.

  • Camera blinding: Camera blinding uses extra light to deny visual input; a 650 nm laser nearly blinded a tested MobilEye camera.The experiment used a camera installed on a non-automated car.
  • Camera blinding: Camera-blinding defenses include redundant cameras, removable near-infrared-cut filters, and predictive frame analysis.Redundancy raises attacker effort but increases implementation cost, while the filter and predictive method require further validation.
  • Adversarial images: Adversarial-image attacks apply small image perturbations that cause CAV vision models to generate incorrect predictions.Potential consequences include ignored alerts or incorrectly read speed limits.
  • Defense criteria: Camera defenses are evaluated against low cost, broad generalization across perturbations or wavelengths, and computational efficiency.These criteria reflect the need to support real-time object detection and varied attack conditions.
  • Adversarial images: Defenses against adversarial images include input preprocessing, adversarial training, and runtime detection of abnormal inputs.The survey calls for comparison based on generalization and computational efficiency.

H. Attacks on communication mechanisms

Communication attacks can falsify information or disrupt V2V and V2I operations, while defenses emphasize authentication and secured network architectures. Existing approaches face real-time, cost, scalability, and realistic-testing constraints.

  • Attack models: Falsified-information attacks use impersonation or Sybil techniques to disrupt CAV operations and traffic flow through V2V and V2I communications.Impersonation can exploit weak authentication by stealing legitimate vehicle identities and broadcasting false information.
  • Defense strategies: Strong authentication, including public-key methods and neighboring-vehicle schemes, is proposed to mitigate falsified-information attacks.Neighbor-based authentication exchanges beacon-derived neighbor records to identify abnormal vehicles, but may fail when too few vehicles are available.
  • Challenges: Authentication defenses must be low-cost, computationally efficient, and capable of real-time operation, while public-key infrastructures may be expensive and too slow for real-time authentication.The survey identifies infrastructure cost and latency as continuing challenges for large-region deployment.
  • Attack models: DoS and DDoS attacks target communication mechanisms and can confuse CAV operations or disrupt traffic flows, potentially across large regions through V2I networks.The cited studies identify vulnerabilities in VANETs and V2I networks.
  • Defense strategies: DoS and DDoS defenses require secured V2V or V2I architectures and protocols, including strategies adapted from traditional centralized networks.The survey notes that these defenses have largely been proposed for centralized-network settings.
  • Challenges: DoS and DDoS defense studies remain theoretical or simulation-based, creating a need for experiments in realistic CAV environments.Realistic testing is proposed to assess suitable defenses for V2V and V2I networks.

IV. RESEARCH CHALLENGES, TRENDS, AND OPEN ISSUES

The paper synthesizes challenges and research trends from academic studies, then compares them with industry publications and reports on CAV cybersecurity. This structure supports analysis of both research findings and industrial development.

  • Academic and industrial perspectives: The section first consolidates challenges and research trends identified in the paper’s academic-literature analysis.These observations derive from the attack, defense, and challenge discussions in earlier sections.
  • Academic and industrial perspectives: It then presents official corporate publications and reports related to CAV cybersecurity, focusing on companies involved in CAV development.The industry review is intended to complement the academic observations.

A. Academic Research

Academic research increasingly examines remote attacks and machine-learning-based defenses, while identifying unresolved attack models and insufficient realistic validation. Open issues include secure software updates and decentralized PKI for real-time communication authentication.

  • Research trends: Remote attack models increasingly target sensors, cameras, and communication mechanisms from roadside locations or other vehicles.The survey cites recent attacks that can be performed remotely.
  • Research trends: New defense strategies increasingly use machine learning and anomaly-based intrusion detection, with requirements for real-time response, low expense, and minimal architectural modification.The stated design requirements prioritize deployability alongside detection capability.
  • Open security issues: Software-update mechanisms remain insufficiently secure because attackers can abuse current mechanisms to compromise CAVs.The survey identifies secure ECU software updating as an obvious research need.
  • Open security issues: A decentralized PKI is identified as necessary for improving V2V and V2I security by reducing key-retrieval latency during real-time authentication.Vehicles may encounter many other CAVs quickly and need timely public-key exchange.
  • Research challenges: Some attack models, including recent GPS spoofing methods, lack corresponding defense strategies in the reviewed literature.The survey characterizes these models as significant and realistic threats.
  • Research challenges: Recently developed attack models still need realistic experiments to assess their threat level and effectiveness in CAV environments.Further experiments could clarify their impacts and encourage defense research.
  • Research challenges: Many proposed defenses have only theoretical or unrealistic simulation-based validation and therefore need realistic experiments for validation and selection.The survey links realistic testing to identifying the most suitable strategies.

B. Industrial development

Industrial efforts include a multi-company safety framework, secure development practices, machine-learning guidance, and a CAN Bus security product. However, the survey finds weak alignment between academic findings and market implementations, with no evidence that deployed CAVs address novel attacks.

  • Industry publications: In July 2019, 11 CAV-related companies published “Safety First For Automated Driving” as a proposed guideline for safer and more secure CAV development.The coalition intended to update the whitepaper and promote it as an international development standard.
  • Industry publications: The whitepaper recommends a Secure Development Lifecycle that integrates security into product development and maintenance through training, software-engineering practices, and sustainment.Its development practices include code review and penetration testing.
  • Industry publications: The whitepaper’s machine-learning guidance covers training and testing data, model architecture, evaluation, deployment, and monitoring.The survey notes that the implementation of these approaches was not described.
  • Industrial products: Guardknox claims that its Communication Lockdown approach provides formally verified, deterministic CAN Bus communication with zero false positives and minimal integration.The company also claims no vehicular hardware modification is required.
  • Industry status: CAV security research and implementation remain at an early stage, with no strong connection between academic research and industry implementation.The survey found no evidence that market CAVs have been updated against novel attack models identified by researchers.
  • Motivation: CAV security has attracted increasing interest because connected automation creates vulnerable components and can expose vehicles to attacks.The survey frames these concerns as accompanying the rapid development and public attention surrounding CAVs.
  • Survey contribution: The survey reviews 184 papers, organizes attacks by target components, access requirements, and motives, and identifies countermeasures, research challenges, and future directions.Its stated goal is to support more secure and trustworthy CAVs.
Loading 2007.08041v1…