IncWP: Incremental Waypoint Search

This repository contains the MATLAB code used in the paper for incremental waypoint search, experiment replay, and result analysis for autonomous vessels.

This repository supports three main tasks:

1. Running waypoint-generation experiments.
2. Running the analysis on original or replayed results.
3. Replaying selected experiments from shared zip data.

Approach Overview

The paper proposes IncWP, an incremental multi-objective waypoint search for testing waypoint-guided autonomous vessels. Instead of evolving the full waypoint set at once, the approach fixes previously selected waypoints and only evolves the current waypoint segment. This reduces simulation cost, lowers failure-detection latency, and allows early navigation failures to be explored before the search spends time on later waypoints.

Selection methods are IncWP_KP, IncWP_Unst, IncWP_Prox, IncWP_Rnd, RandomSearch, IncWP_Kmeans, and FullWP. Supported vessel names are remus100, nspauv, and mariner.

Setup

Clone this repository:

git clone https://github.com/Karolinenylaender/WPgen-extended.git

Clone the external frameworks used by the project:

git clone https://github.com/cybergalactic/MSS.git
git clone https://github.com/BIMK/PlatEMO.git BIMK-PlatEMO-4.7.0.0

The adapted NSGA-II-Adapted code used by this project is tracked in this repository under frameworks/evolutionaryPlatform/NSGA-II-Adapted.

Dependencies and licensing:

• MSS by Thor I. Fossen and contributors
• PlatEMO by the BIMK Group
• the adapted NSGA-II-Adapted code used in this repository

Place the downloaded folders either:

• directly in the repository root as MSS, BIMK-PlatEMO-4.7.0.0, and NSGA-II-Adapted, or
• already inside frameworks/MSS and frameworks/evolutionaryPlatform

Open MATLAB in the repository root and run:

setupProject

setupProject adds the required MATLAB paths, moves the framework folders into frameworks/ if needed, and creates the experimentsData, replicationRuns/experiments, and replicationData folder structure used by the project.

If the framework folders are placed in the repository root, setupProject moves them into:

• frameworks/MSS
• frameworks/evolutionaryPlatform/BIMK-PlatEMO-4.7.0.0
• frameworks/evolutionaryPlatform/NSGA-II-Adapted

Running Experiments

The main experiment entry script is runExperiments.m. The approach implementations are in scripts/approaches. After setupProject, the repository is ready to run and resultsPath should point to experimentsData.

To run one standard incremental waypoint experiment directly:

vesselName = "remus100";
selectionType = "IncWP_KP";
experimentNumber = 1;
numGenerations = 1000;
populationSize = 10;
resultsPath = "experimentsData";

runIncWP(vesselName, selectionType, resultsPath, experimentNumber, numGenerations, populationSize)

Important options:

• vesselName: remus100, nspauv, or mariner.
• selectionType: IncWP_KP, IncWP_Unst, IncWP_Prox, IncWP_Rnd, or IncWP_Kmeans.
• resultsPath: explicit output folder path, normally experimentsData.
• experimentNumber: experiment identifier.
• numGenerations: default is 1000.
• populationSize: default is 10.

Approach Variants

The table below links the paper naming, MATLAB naming, and the script used to run each approach in scripts/approaches.

Paper name	MATLAB selection string	Run function	Meaning
IncWP_KP	IncWP_KP	runIncWP(..., "IncWP_KP", ...)	Selects the knee point of the Pareto front, representing a compromise between proximity to the original waypoint and path instability.
IncWP_Unst	IncWP_Unst	runIncWP(..., "IncWP_Unst", ...)	Selects the individual with the highest instability objective value to prioritize failure-revealing paths.
IncWP_Prox	IncWP_Prox	runIncWP(..., "IncWP_Prox", ...)	Selects the individual closest to the original waypoint, prioritizing realistic and minimal waypoint changes.
IncWP_Rnd	IncWP_Rnd	runIncWP(..., "IncWP_Rnd", ...)	Selects one reachable Pareto-front individual at random.
IncWP_Kmeans	IncWP_Kmeans	runIncWPKmeans(...)	Selects multiple starting waypoints using clustering over the final population.
RandomSearch	RandomSearch	runRandomSearch(...)	Random incremental baseline without guided evolutionary selection.
FullWP	FullWP	runFullWP(...)	Full waypoint-set search that evolves the complete waypoint sequence at once.

The first five rows are the IncWP family. RandomSearch and FullWP are the baselines used in the paper.

Code Structure

scripts
Main experiment code. This folder contains the approach entry points in approaches/, the search and simulation logic in vesselSearch/, and the top-level experiment runner in runExperiments.m.

Important subfolders:

• approaches: entry points for IncWP, IncWP_Kmeans, RandomSearch, and FullWP
• vesselSearch/incrementalSearch: incremental waypoint search and path simulation
• vesselSearch/globalSearch: full waypoint-set search and evaluation code
• vesselSearch/helpers: shared helper functions used by the search code

analysis
Analysis pipeline for processing experiment outputs and reproducing the reported metrics and figures.

Important subfolders:

• dataProcessing: loads raw experiment outputs and restructures them for later analysis
• metric: computes the reported metrics and statistical summaries
• displayFunctions: plotting and result-display helpers used by the analysis code

replication
Scripts for unpacking the tracked replication data and rerunning the missing simulation outputs.

replicationData
Tracked publication data, including the packaged experiment zip files and zipped analysed results.

frameworks
External frameworks used by the project, including the adapted evolutionary platform and MSS.

Analysis

Use runAnalysis.m to analyse either original experiment data or replayed data.

For the original experiment data:

runAnalysis("remus100", "experimentsData")

For replayed data:

runAnalysis("remus100", "replicationRuns/experiments")

Replication

Use the replication workflow only when you want to rerun the packaged experiment data. The tracked experiment zip files are stored under replicationData/zippedExperiments/. The path-simulation files are too large to include directly, so runReplication unzips the packaged data and reruns those simulation outputs locally.

To rerun the replication data:

runReplication

runReplication imports the tracked experiment zip files, reruns the simulations in replicationRuns/experiments/, and then runs the analysis on the replayed outputs.

Dependencies

This project uses third-party code from:

• PlatEMO, developed by Ye Tian, Ran Cheng, Xingyi Zhang, Yaochu Jin, and the BIMK Group
• MSS (Marine Systems Simulator), developed by Thor I. Fossen and contributors
• knee_pt, developed by Dmitry Kaplan

We do not own those components. Keep their license files, notices, and requested citations when redistributing or publishing from this repository.