UAVCAN Attack Dataset 2026 (LUMI)

1. Dataset Overview

• UAVCAN Protocol: UAVCAN v0 (now maintained as DroneCAN) is an application-layer protocol running over CAN bus, designed for reliable, realtime distributed communication in Unmanned Aerial Vehicles (UAVs), robotics, and aerospace systems.

• Dataset Purpose: Built to address key structural limitations in existing UAVCAN datasets—such as single-attack isolation and non-aligned

Transfer IDs (TIDs)—by providing highly complex, compound intrusions and protocol-aware threat scenarios.

• Data Collection Method: Collected from a physical UAV operating under normal flight conditions while executing real-time attack injections targeted at high-impact motor control parameters.

2. Testbed Environment & Hardware

• Target Vehicle: Commercial UAV platform driven by a Pixhawk 4 flight controller connected to four UAVCAN-compatible Electronic Speed Controllers (ESCs).

• Malicious Node Setup: A Raspberry Pi 4 Model B equipped with a CAN shield attached directly to the UAVCAN bus interface as a terminal node.

• Control Software: Real-time attack packets generated using the python-can library over SSH connection.

3. Dataset Statistics

Both 'Attack' and 'Normal' labels are defined at the frame level.

4. Key Features & Attack Scenarios

Core Attack Characteristics

• Watchdog-Mimicking Attacks: Attack messages mimic the legitimate cyclic Transfer ID (TID) sequence (031) in the UAVCAN tail byte, making simple sequence-based detection ineffective.

• Compound & Concurrent Attacks: Includes complex intrusion scenarios where multiple attack vectors are executed simultaneously or sequentially.

• Flight-Critical Targets: Intentionally targets motor control IDs to directly challenge the flight stability and functional safety of the UAV.

Attack Types Included

1. Flooding: High-frequency message injection disrupting bus availability.

2. Fuzzing: Arbitrary and randomized payload injections targeting motor control channels.

3. Replay: Retransmission of valid historical flight traffic.

4. Compound Injections: Multi-threaded combinations including Flooding, Fuzzing, and Replay executed simultaneously.

5. Reference & Citation

If you use this dataset in your research, please cite the following paper:

@INPROCEEDINGS{LUMI_Song, author={Song, Yuchan and Kim, Huy Kang}, booktitle={2026 IEEE Intelligent Vehicles Symposium (I

V)},

title={LUMI: Lightweight UAVCAN Message Window-Based Intrusion Detection System},

year={2026},

volume={},

number={},

pages={2102-2109},

keywords={Modeling;Timing;Controller area networks;Signal detection;Windows;Printing;Autonomous aerial vehicles;Entropy;Security;Protocols;intrusion detection system (IDS);rule-based IDS;lightweight security;unmanned aerial vehicle (UAV);CAN protocol;UAVCAN protocol}, doi={10.1109/IV66570.2026.11623974}}

6. DATASET DOWNLOAD

Download Link: Download

7. CONTACT

Yuchan Song (thddbcks618@korea.ac.kr) or Huy Kang Kim (cenda@korea.ac.kr)