jonasbrondum / cyber-resilient-marine-navigation

Selected code from my M.Sc. thesis: Cyber-resilient multi-modal sensor fusion for autonomous ship navigation

Cyber-resilient multi-modal sensor fusion for autonomous shipnavigation

This repository is a selection of code used in my master thesis with the title: Cyber-resilient multi-modal sensor fusion for autonomous shipnavigation. I've tried to clean the code and make it as readable as possible. Hopefully, this can be useful for anyone wanting to implement their own error-state kalman filter.

Project abstract

Ensuring the integrity of navigation sensors is paramount for safe maritime navigation. Un- fortunately, cyber-attacks targeting satellite positioning information and onboard sensors have exposed vulnerabilities, posing a significant threat to navigation safety. In near-coast and urban environments with high traffic density and static structures, the risk of collisions and grounding is elevated. Without a navigator’s sanity check of crucial information such as position, heading, and velocity, cyber attacks could lead to accidents by compromising accurate data.

This thesis addresses the challenge of enhancing the cyber-resilience of maritime surface vessel navigation systems. Various sensor systems are evaluated, and auxiliary sensing methods are proposed to create a GNSS-independent navigation system. The thesis introduces two methods—point cloud registration and template matching—for an auxiliary positioning system. Additionally, an observer is designed to leverage the existing array of sensors within a framework supported by auxiliary measurements.

The developed navigation system is tested using real-world data collected in Copen- hagen’s inner harbor. Despite the absence of GNSS measurements, the system pro- duces sufficiently accurate positioning. The resulting position estimate demonstrates the cyber-resilience of the proposed solution for GNSS-independent navigation.

Contents

The project consisted of several methods for calculating an odometry to use in localization.

1. LiDAR odometry from point cloud registration. This is located in lidar_odometry.py.
2. Template matching (TM) odometry using LiDAR with electronic navigational charts (ENC) or open-street map (OSM). This is located in tm_odometry.py.

These odometries were along with GPS, Compass, and IMU measurements used in the error-state Kalman filter for continuous localization. The error-state kalman filter is implemented as a class in eskf.py.

The pipeline is combined in pipeline.py

Dataset

The pipeline as it is implemented here makes use of pickled DataFrames. The data used for the project is not for public use, so it is, unfortunately, unavailable for me to share. The loadData()in data_samples.py can easily be modified to fit other data storage.

The dataset was partitioned into smaller sets for development such that the relative East-North-Up coordinate system could be utilized.

The current structure accomodates 5 DataFrames:

lidar_df

timestamp	path
Unix Epoch format	path to .pcd-files

gps_df

timestamp	latitude	longitude	East	North	sog	cog
in Unix Epoch format	-	-	-	-	speed over ground	course over ground

imu_df

timestamp	q0	q1	q2	q3	ax	ay	az	gx	gy	gz
in Unix Epoch format	quaternions	-	-	-	acceleration	-	-	angular velocities	-	-

heading_df

timestamp	heading uncorrected	heading vessel
in Unix Epoch format	-	-

enc_df

timestamp	path	enc_max_range	latitude	longitude	pxwidth
in Unix Epoch format	path to enc-generated images	max range in meters	center of image	center of image	pixelwidth of the image

How-to

Project requirements can be installed from requirements.txt.

pipeline.py is set up to run with the project parameters.

Note: The parameters used for the different sections of the pipeline are mostly defined in the parameters.pyclass with the call of loadParameters().

Quality statement

This is research quality code, not suitable for production or any real use case. The code is provided as-is with no guarantees or warranties.