Baseline Implementations for Safe Formation Control of Networked Wheeled Robots

This repository contains a reproducible four-method MATLAB simulation for comparing formation accuracy and obstacle safety in a team of eight unicycle-type wheeled robots.

Compared methods

ID	Repository name	Learning	Identifier	Prediction	Obstacle treatment
M1	Learning-based formation baseline	Yes	No	No	None
M2	Tightened sampled-action predictive baseline	No	No	Yes	Predicted-distance feasibility test and hard backup
M3	CLF--LBF-inspired formation baseline	No	No	No	LBF-inspired obstacle repulsion
M4	Proposed identifier-assisted safe learning controller	Yes	Yes	No in this script	Barrier penalty and barrier control correction

The three baselines are adaptations of methodological ideas in the cited literature. They are not claimed to be line-by-line reproductions of the published algorithms.

Repository structure

.
├── README.md
├── CITATION.cff
├── references.bib
├── src/
│   ├── four_method_comparison.m
│   └── four_method_comparison_clean.m
├── docs/
│   ├── COMPARISON_METHODS.md
│   ├── PARAMETERS.md
│   ├── METRICS_AND_FAIRNESS.md
│   ├── CODE_TO_PAPER_MAPPING.md
│   ├── PAPER_BASELINE_DESCRIPTION.tex
│   ├── RELEASE_CHECKLIST.md
│   └── CLEANUP_DIFF.md
└── results/
    └── README.md

src/four_method_comparison.m is an exact MATLAB copy of the supplied script.

src/four_method_comparison_clean.m changes only stale comments and obsolete vertical plot markers. Controller equations, numerical parameters, disturbances, metrics, and integration are unchanged.

Requirements

• MATLAB R2020a or later is recommended because the script uses exportgraphics.
• No additional MATLAB toolbox is required by the supplied implementation.
• The script uses local functions in a script file.

Running the simulation

cd src
four_method_comparison_clean

The script runs all methods with identical initial conditions and deterministic disturbance sequences.

Generated outputs

comparison_metrics.csv
comparison_timeseries.csv
comparison_results.mat
Fig1_Trajectories_M1.png
Fig1_Trajectories_M2.png
Fig1_Trajectories_M3.png
Fig1_Trajectories_M4.png
Fig2_FormationErrorComparison.png
Fig3_SafetyMarginComparison.png
Fig4_RMSE_Bar.png
Fig5_MaxError_Bar.png
Fig6_SafetyMargin_Bar.png
Fig7_DisturbanceProfiles.png

Move the reviewed final outputs to results/ before creating a release.

Reproducibility statement

All methods use the same robot model, leader trajectory, desired formation, communication graph, initial-condition random seed, obstacle geometry, deterministic disturbances, input limits, integration settings, and metric definitions.

Exact controller equations and implementation differences are documented in docs/COMPARISON_METHODS.md. Complete numerical settings are listed in docs/PARAMETERS.md.

Terminology limitations

The current M2 code is a local sampled-action finite-horizon predictive controller. It does not exchange predicted trajectories or solve coupled local optimal-control problems; it should not be described as an exact cooperative DMPC implementation.

The current M3 code is CLF--LBF-inspired. Formal CLF decrease and LBF invariance should only be claimed if separately derived and proved.

The current M4 code contains actor--critic-style learning, an online residual identifier, and a barrier safety mechanism. It has no finite prediction horizon in this script.

Citation and release

Before publication:

1. replace all placeholders in CITATION.cff;
2. add a software license;
3. create a versioned GitHub release;
4. archive the release using a DOI-granting repository;
5. cite the fixed release or DOI rather than the mutable default branch.