Slamspoof

Codes for (ICRA'25) "SLAMSpoof: Practical LiDAR Spoofing Attacks on Localization Systems Guided by Scan Matching Vulnerability Analysis"

Paper Link

"SLAMSpoof: Practical LiDAR Spoofing Attacks on Localization Systems Guided by Scan Matching Vulnerability Analysis", Rokuto Nagata, Kenji Koide, Yuki Hayakawa, Ryo Suzuki, Kazuma Ikeda, Ozora Sako, Qi Alfred Chen, Takami Sato, and Kentaro Yoshioka

What is Slamspoof ?

Slamspoof is a framework that analyzes LiDAR SLAM and reveals vulnerabilities to adversarial laser pulse injections, or LiDAR spoofing. The vulnerabilities are quantitatively evaluated using the SMVS (Scan Matching Vulnerability Score) metric, and the values can be referenced across the entire map. Please note that this repository focuses on localization using LiDAR-based localization (Visual Odometry is not covered, sorry). Additionally, the framework supports kiss-icp and A-LOAM as localization methods.

Related repositories

• evo
  • The output results are generated in a format that can be evaluated using evo.
• A-LOAM
  • The “main” branch supports KISS-ICP, the A-LOAM version is available in the “loam” branch

Prerequisites

• ROS noetic

  • Installation is here (ROS wiki)
  • Workspace should be built catkin build.

    sudo apt-get install python3-catkin-tools
    source /opt/ros/noetic/setup.bash
    mkdir -p ~/your_ws/src
    cd ~/your_ws
    catkin build
    source devel/setup.bash
    

• kiss-icp

  • Install here

• small_gicp

  • Install here. We recommend installation from source.

• Python packages

  • numpy
  • pandas
  • matplotlib
  • rosbags

Install Slamspoof

First, confirm prerequisites are correctly installed.

# for ROS noetic
cd your_ros_workspace/src
git clone https://github.com/ngtroku/slamspoof.git
cd .. && catkin build 
source devel/setup.bash

How to use

• This repository has two functions
  • Calculate SMVS and visualization
  • Simulation of LiDAR Spoofing Attacks

How to run SLAM and calculate SMVS

roslaunch slamspoof run_slam.launch

And run rosbag in another terminal

rosbag play (your_rosbag).bag

Settings

• Line 3,4,5: Directory path to save simulation results.
  • Line 3: name_smvs_dir is path for storing SMVS data
  • Line 4: name_vul_dir is path for Vulnerable direction of SLAM system
  • Line 5: name_traj_dir is path storing SLAM results
• Line 8 subscribe_topic_name : Name of point cloud rostopic to input into SLAM system
• Line 15 lidar_topic_length : PointCloud topic format
  • If you use Velodyne LiDAR, set value 22
  • If you use Livox LiDAR, set value 18
  • If you use Pointcloud topic edited rosbag_editer.launch, set value 12

How to visualize SMVS

roslaunch slamspoof visualizer.launch

Setting

• Line2 smvs_file_path: Directory of SMVS data

How to simulate LiDAR Spoofing Attacks

Normal Trajectory must be obtained before simulating attack.

roslaunch slamspoof rosbag_editer.launch

Simulation settings are changed from config.json

Settings

• Line3 input_file : rosbag path used for simulation.
• Line4 output_file : Output destination for simulation results.
• Line5 spoofing_mode : type of spoofing, if "removal", HFR attack, if "static" inject false wall.
• Line6, 7 spoofer_x and spoofer_y : Spoofer placement position. (specified by x and y coordinates)
• Line8 distance_threshold : Distance at which Spoofer can track and attack. Spoofing occurs when the target enters within a radius of distance_threshold meters from the attacking device.
• Line9 lidar_topic : Input rosbag pointcloud topic name.
  • Line10 lidar_topic_length : If you use Velodyne LiDAR, set value 22. If you use Livox LiDAR, set value 18.
• Line12 spoofed_topic : Topic name of the simulated PointCloud. It is recommended not to change from the original Topic name.
• Line13 spoofing_range : The horizontal azimuthal range of the pointcloud that Spoofer can tamper with in a single frame.
• Line14 wall_dist : Used only in static mode. How far away from the target the wall is injected. (Unit : meter)
• Line17 horizontal_resolution : Horizontal resolution of LiDAR.
• Line18 vertical_lines: Number of LiDAR vertical scan lines.
• Line19 spoofing_rate : Values between 0.0 and 1.0. The higher the value, the more random noise is injected during the HFR attack. Default values are experimental and recommended ones.
• Line22 minimum_measureing_distance : The simulation truncates the pointcloud below a certain distance from LiDAR during simulation.
• Line23 maximum_measureing_distance : The simulation truncates the pointcloud beyond a certain distance from LiDAR during the simulation.
• Line24 minimum_height_htreshold : Eliminate points below a certain height during simulation to remove the influence of the ground pointcloud.