AMCFF-RL: An Adaptive Multi-Modal CAN Bus Fuzzing Framework Leveraging Deep Reinforcement Learning

Manu Jo Varghese; Frank Jiang; Abdur Rakib; Robin Doss; Adnan Anwar

doi:10.1109/OJVT.2026.3659052

AMCFF-RL: An Adaptive Multi-Modal CAN Bus Fuzzing Framework Leveraging Deep Reinforcement Learning

Manu Jo Varghese
, Frank Jiang
, Abdur Rakib
, Robin Doss
, Adnan Anwar

Centre for Future Transport and Cities

Deakin University

Research output: Contribution to journal › Article › peer-review

Abstract

The increasing complexity and connectivity of modern vehicles have made automotive networks, particularly the Controller Area Network (CAN) bus, vulnerable to cyberattacks. Fuzzing is a critical technique for proactively finding security weaknesses, but traditional methods are inefficient and struggle to scale with the complexity of modern vehicles. This paper introduces AMCFF-RL, an
adaptive framework that uses Deep Reinforcement Learning (DRL) with multi-modal feature extraction to systematically analyse for vulnerabilities. Rather than relying on unguided or purely random fuzzing, AMCFF-RL integrates multi-modal feature extraction with DRL and advanced visualization, allowing
it to learn and adapt its strategy based on real-time feedback from the network and thereby improve the efficiency and effectiveness of the fuzzing process. Comprehensive visualization tools serve a dual purpose: they offer human-interpretable insights while also generating rich feature representations that support the anomaly detection pipeline and the DRL agent.

Original language	English
Pages (from-to)	612-625
Number of pages	14
Journal	IEEE Open Journal of Vehicular Technology
Volume	7
DOIs	https://doi.org/10.1109/OJVT.2026.3659052
Publication status	Accepted/In press - 20 Jan 2026

Bibliographical note

©2026 The Authors.
This is an Open Access article distributed under the terms of the Creative
Commons Attribution License (http://creativecommons.org/licenses/by/4.0/)
Under this licence, users are permitted to share, download, copy, and redistribute the material in any medium or format, and—where applicable—adapt or build upon the work, provided they comply with the conditions of the stated licence

Keywords

Automotive security
CAN Bus
Deep Reinforcement Learning
Fuzzing
Multi-modal Feature Extraction
Vulnerability Detection
Cyber-Physical Systems

Access to Document

10.1109/OJVT.2026.3659052Licence: CC BY

Varghese2026VoRFinal published version, 2.51 MBLicence: CC BY

Cite this

@article{9f5fcf6a6ab5489fb18ce8d6c55e9c7b,

title = "AMCFF-RL: An Adaptive Multi-Modal CAN Bus Fuzzing Framework Leveraging Deep Reinforcement Learning",

abstract = " The increasing complexity and connectivity of modern vehicles have made automotive networks, particularly the Controller Area Network (CAN) bus, vulnerable to cyberattacks. Fuzzing is a critical technique for proactively finding security weaknesses, but traditional methods are inefficient and struggle to scale with the complexity of modern vehicles. This paper introduces AMCFF-RL, an adaptive framework that uses Deep Reinforcement Learning (DRL) with multi-modal feature extraction to systematically analyse for vulnerabilities. Rather than relying on unguided or purely random fuzzing, AMCFF-RL integrates multi-modal feature extraction with DRL and advanced visualization, allowing it to learn and adapt its strategy based on real-time feedback from the network and thereby improve the efficiency and effectiveness of the fuzzing process. Comprehensive visualization tools serve a dual purpose: they offer human-interpretable insights while also generating rich feature representations that support the anomaly detection pipeline and the DRL agent.",

keywords = "Automotive security, CAN Bus, Deep Reinforcement Learning, Fuzzing, Multi-modal Feature Extraction, Vulnerability Detection, Cyber-Physical Systems",

author = "Varghese, \{Manu Jo\} and Frank Jiang and Abdur Rakib and Robin Doss and Adnan Anwar",

note = "{\textcopyright}2026 The Authors. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/) Under this licence, users are permitted to share, download, copy, and redistribute the material in any medium or format, and—where applicable—adapt or build upon the work, provided they comply with the conditions of the stated licence",

year = "2026",

month = jan,

day = "20",

doi = "10.1109/OJVT.2026.3659052",

language = "English",

volume = "7",

pages = "612--625",

journal = " IEEE Open Journal of Vehicular Technology",

issn = "2644-1330",

publisher = "IEEE",

}

TY - JOUR

T1 - AMCFF-RL

T2 - An Adaptive Multi-Modal CAN Bus Fuzzing Framework Leveraging Deep Reinforcement Learning

AU - Varghese, Manu Jo

AU - Jiang, Frank

AU - Rakib, Abdur

AU - Doss, Robin

AU - Anwar, Adnan

N1 - ©2026 The Authors. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/) Under this licence, users are permitted to share, download, copy, and redistribute the material in any medium or format, and—where applicable—adapt or build upon the work, provided they comply with the conditions of the stated licence

PY - 2026/1/20

Y1 - 2026/1/20

N2 - The increasing complexity and connectivity of modern vehicles have made automotive networks, particularly the Controller Area Network (CAN) bus, vulnerable to cyberattacks. Fuzzing is a critical technique for proactively finding security weaknesses, but traditional methods are inefficient and struggle to scale with the complexity of modern vehicles. This paper introduces AMCFF-RL, an adaptive framework that uses Deep Reinforcement Learning (DRL) with multi-modal feature extraction to systematically analyse for vulnerabilities. Rather than relying on unguided or purely random fuzzing, AMCFF-RL integrates multi-modal feature extraction with DRL and advanced visualization, allowing it to learn and adapt its strategy based on real-time feedback from the network and thereby improve the efficiency and effectiveness of the fuzzing process. Comprehensive visualization tools serve a dual purpose: they offer human-interpretable insights while also generating rich feature representations that support the anomaly detection pipeline and the DRL agent.

AB - The increasing complexity and connectivity of modern vehicles have made automotive networks, particularly the Controller Area Network (CAN) bus, vulnerable to cyberattacks. Fuzzing is a critical technique for proactively finding security weaknesses, but traditional methods are inefficient and struggle to scale with the complexity of modern vehicles. This paper introduces AMCFF-RL, an adaptive framework that uses Deep Reinforcement Learning (DRL) with multi-modal feature extraction to systematically analyse for vulnerabilities. Rather than relying on unguided or purely random fuzzing, AMCFF-RL integrates multi-modal feature extraction with DRL and advanced visualization, allowing it to learn and adapt its strategy based on real-time feedback from the network and thereby improve the efficiency and effectiveness of the fuzzing process. Comprehensive visualization tools serve a dual purpose: they offer human-interpretable insights while also generating rich feature representations that support the anomaly detection pipeline and the DRL agent.

KW - Automotive security

KW - CAN Bus

KW - Deep Reinforcement Learning

KW - Fuzzing

KW - Multi-modal Feature Extraction

KW - Vulnerability Detection

KW - Cyber-Physical Systems

U2 - 10.1109/OJVT.2026.3659052

DO - 10.1109/OJVT.2026.3659052

M3 - Article

SN - 2644-1330

VL - 7

SP - 612

EP - 625

JO - IEEE Open Journal of Vehicular Technology

JF - IEEE Open Journal of Vehicular Technology

ER -

AMCFF-RL: An Adaptive Multi-Modal CAN Bus Fuzzing Framework Leveraging Deep Reinforcement Learning

Abstract

Bibliographical note

Keywords

Access to Document

Fingerprint

Cite this