1.Key Laboratory of Aerospace Information Security and Trusted Computing,Ministry of Education,School of Cyber Science and Engineering,Wuhan University,Wuhan 430072,Hubei,China
2.The State Key Laboratory of Blockchain and Data Security,Zhejiang University,Hangzhou 310058,Zhejiang,China
Existing coverage-guided fuzzing tools in intelligent unmanned systems face issues such as efficiency degradation and uncontrollable states. To address this problem, this paper selects the PX4 open-source unmanned system as the research object. Firstly, it analyzes three core causes of performance degradation in fuzzing: asynchronous thread execution, internal and external signal interaction, and spatiotemporal state residue. Based on these three core causes, targeted mitigation mechanisms are designed, including forced synchronization, message filtering, time variable substitution and improved selective instrumentation, to generate test-specific firmware. Additionally, a fuzzing framework for intelligent unmanned systems, IUSFUZZ, is constructed. Compared with the mainstream fuzzing tool AFL++, IUSFUZZ has improved stability by 16.5 percentage points. To discover the same vulnerabilities, AFL++ requires 7.5 times as long as IUSFUZZ. IUSFUZZ successfully uncovered a zero-day vulnerability triggered through a network channel and obtained a CVE number.
随着智能无人系统(如无人机、无人车及服务机器人等)在工业自动化、物流配送、环境监测及消费娱乐等领域的广泛应用[1-3],其对外部环境的高度感知能力与自主决策机制使其能够执行日益复杂的任务,但这种高度互联与功能集成的特性也显著扩大了系统的攻击面,因为智能无人系统需要首先对各类数据(如传感器输入、用户指令和自身状态等)进行解析和判断才能确定后续执行的动作,错误或恶意的数据都可能影响智能无人系统的决策过程,引发安全问题。如通过伪造胎压监测系统传感器数据触发整数溢出漏洞,即可在无线范围内对无人车进行远程任意代码执行(CVE-2025-2082);又如通过向目标机器人发送精心构造的网络数据包,即可使机器人进入安全停止状态,从而导致生产中断(CVE-2024-6157)。无人系统控制器承担着状态感知、姿态解算、导航控制及执行机构驱动等关键功能,是保障系统安全稳定运行的核心模块,该模块需实时处理来自地面控制站(Ground Control Station, GCS)、遥控器、GPS模块及多种机载传感器的多源异构输入信号,形成了多维度、高并发的通信接口体系,从而引入了诸多潜在入侵途径。值得注意的是,当前主流开源无人系统广泛采用MAVLink协议(Micro Air Vehicle Link)进行跨组件通信。该协议设计之初以轻量化和低延迟为核心目标,缺乏身份认证、数据完整性校验和加密传输等基本安全机制[4],导致其极易遭受中间人攻击(Man-in-the-Middle Attack)、数据注入、指令篡改及远程控制劫持等恶意行为。更为严峻的是,研究表明,PX4-Autopilot[5]和Ardupilot[6]等开源无人系统占全球民用及行业级市场份额超过30%[7],这意味着一旦核心协议或软件存在可利用漏洞,将可能引发大规模的安全风险事件。
为验证IUSFUZZ的漏洞挖掘效率,本文将其与针对PX4控制程序MAVLink接口的黑盒模糊测试工具4DFuzzer和主流模糊测试工具AFL++进行了对比实验。值得注意的是,为了实时捕获触发拒绝服务(Deny of Service,DoS)类漏洞的测试用例,本文在PX4控制程序的关机请求函数中插入了一段检测代码,即控制程序在解析MAVLink数据包后尝试进行关机时,检测代码就会通知内存检测工具(如Address Sanitizer)触发警报,以便测试工具及时进行响应,防止其因控制程序的延时命令执行机制而导致关键数据丢失。测试漏洞的信息如下:
1) 在SITL仿真环境中,攻击者向目标机器控制程序的任意MAVLink端口持续发送恶意数据包,攻击目标在遭受攻击后会在日志系统中输出低电量警报“Dangerously low battery! Shutting system down in 60 seconds”,并会在1 min后直接关机,如果目标机器控制程序在遭受攻击前连接了QGC(QGroundControl)地面站程序[24],在攻击者实施攻击后将会看到QGC状态栏出现多个低电量图标并直接与目标机器控制程序断开连接。
a. 处于等待启动状态。受害者尝试使用QGC向目标机器发送启动命令,目标机器将因为无法通过安全检查(低电量警报)而拒绝启动。
b. 处于运行状态。如图9所示,受害者将观察到QGC状态栏显示了一个低电量图标警报,状态栏变为红色,同时受害者无法再利用QGC控制目标机器。若目标机器为无人机,那么使用遥控器可以勉强控制目标无人机的水平方向,但无法控制目标无人机的垂直升降,目标无人机的运动轨迹呈现随机变化趋势。当攻击者停止攻击后,无人机将立刻进入低电量紧急状态并采取安全措施,开始垂直降落并准备着陆。
CompanyAlphabet.Wing autonomous, lightweitght delivery drones to transport goods[EB/OL]. [2025-08-07].
[2]
MÜLLERC. World robotics 2024 – service robots[EB/OL].[2025-08-07].
[3]
SINGHI. Super Bowl weekend: 1000 strong drone show to light up Vegas sky[EB/OL]. [2024-02-08].
[4]
KOUBÂAA, ALLOUCHA, ALAJLANM, et al. Micro air vehicle link (MAVlink) in a nutshell: A survey[J]. IEEE Access, 2019, 7: 87658-87680. DOI:10.1109/ACCESS.2019.2924410 .
[5]
MEIERL, AGARD, KÜNGB, et al. PX4-Autopilot[EB/OL]. [2025-08-07].
[6]
PRATAPA, BONNEYB, PUROHITS B, et al. Ardupilot[EB/OL]. [2025-08-07].
[7]
BENOWITZD.The rise of open-source drones[EB/OL].[2021-05-30].
[8]
MALVIYAV K, MINNW, SHARL K, et al. Fuzzing drones for anomaly detection: A systematic literature review[J]. Computers & Security, 2025, 148: 104157. DOI:10.1016/j.cose.2024.104157 .
[9]
CHENH, GUOS, XUEY, et al. MUZZ: Thread-aware grey-box fuzzing for effective bug hunting in multithreaded programs[EB/OL]. [2025-03-10].
[10]
LIJ Q, ZHOUY J, WUL. Improving multitasking DBMS fuzzing with more accurate coverage and testcase trimming[J]. IEEE Transactions on Dependable and Secure Computing, 2025, 22(3): 2756-2770. DOI:10.1109/TDSC.2024.3521591 .
WANGL N, ZHAOJ C, YUR W, et al. Research on fuzz testing method for ARM secure monitor[J]. Journal of Wuhan University (Natural Science Edition), 2024, 70(4): 453-462. DOI:10.14188/j.1671-8836.2023.0009(Ch ).
[13]
CHANGZ W, ZHANGH F, YANGY, et al. Fuzzing drone control system configurations based on quality-diversity enhanced genetic algorithm[C]//Artificial Intelligence Security and Privacy. Singapore: Springer, 2024: 499-512. DOI:10.1007/978-981-99-9785-5_35 .
[14]
KIMT, KIMC H, RHEEJ J, et al. RVFuzzer: Finding input validation bugs in robotic vehicles through control-guided testing[EB/OL]. [2025-03-12]. DOI: 10.1145/3548606.3560558 .
[15]
LIG P, LIY R, JHA S, et al. AV-FUZZER: Finding safety violations in autonomous driving systems[C]//2020 IEEE 31st International Symposium on Software Reliability Engineering (ISSRE). New York: IEEE Press, 2020: 25-36. DOI:10.1109/issre5003.2020.00012 .
[16]
WANGY, YANGC, ZHANGX D, et al. DPFuzzer: Discovering safety critical vulnerabilities for drone path planners[C]//2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE). New York: IEEE Press, 2025: 1973-1985. DOI:10.1109/ICSE55347.2025.00051 .
[17]
DOMINK, MARINE, SYMEONIDISI. Security analysis of the drone communication protocol: Fuzzing the MAVLink protocol[EB/OL].[2025-08-07].
[18]
BOB4Drone.4D-Fuzzer[EB/OL].[2025-08-07].
[19]
KIMH, OZMENM O, BIANCHIA, et al. PGFUZZ: Policy-guided fuzzing for robotic vehicles[EB/OL]. [2025-09-27]. DOI: 10.14722/ndss.2021.24096 .
[20]
MEIERL, AGARD, KÜNGB, et al. Simulation[EB/OL].[2025-08-07].
[21]
FIORALDIA, MAIERD, EIßFELDTH, et al. AFL++: Combining incremental steps of fuzzing research[EB/OL]. [2025-09-28].
[22]
DROBILLAD R. Llvm_mode persistent mode[EB/OL].[2025-08-07].
[23]
MEIERL, HONEGGERD, POLLEFEYSM. PX4: A node-based multithreaded open source robotics framework for deeply embedded platforms[C]//2015 IEEE International Conference on Robotics and Automation (ICRA). New York: IEEE Press, 2015: 6235-6240. DOI:10.1109/ICRA.2015.7140074 .
[24]
VANHAUSER-THC. Using AFL++ with partial instrumentation[EB/OL].[2025-08-07].