SUPPLEMENTARY MATERIAL
Mode-Aware Adaptive Navigation for Autonomous Underwater Vehicles using
Multimodal Sensing and Optical Feedback in Simulation Environment

C. Alexandris, P. Papageorgas, D. Piromalis
Department of Electrical and Electronics Engineering, University of West Attica


CONTENTS

figures/
    The two figures reproduced in the article, in PDF (vector, as typeset) and
    PNG. fig_mode_aware_architecture is Figure 1; fig_study3_partb_effects is
    Figure 2.

final_tables/
    Every number reported in the article, machine readable.
      contrasts.csv                        paired effects, means and 95% CIs
      means.csv                            per-policy means, both parts
      part_a_family_summary.csv            per-family Part A summary
      adaptation.csv                       adaptation rates and delays
      switching_and_intervention_cost.csv  switching and intervention cost
      safety_and_completion.csv            safety and completion
      final_decision_record.json           predeclared decision rules, evaluated
      study3_final_tables.json             all of the above in one bundle

heldout_final_root36000000/
    The final held-out evaluation of the corrected controller, executed once at
    seed root 36,000,000. This is the evidence behind Section 4 of the article.
      packets/                             2010 immutable result packets
      analysis_partA_scripted.json         Part A, 270 paired scripted cells
      analysis_partB_generated.json        Part B, 400 paired generated environments
    Part A and Part B are separate experimental blocks and are never pooled.

heldout_precorrection_root32000000/
    The earlier held-out block at seed root 32,000,000, reported in Section 4.1.
    It measures the pre-correction controller and is retained unmodified for
    provenance. It must not be combined with the block above: the two measure
    different controllers.
      packets/                             810 immutable result packets
      analysis.json                        analysis of that block

code/
    analyse_heldout_v2.py    reproduces the Part A and Part B analyses
    build_final_tables.py    regenerates every file in final_tables/

supplementary_analyses/
    The two supplementary analyses reported in the revised article: the
    representative simulation case of Section 4.4 and Figure 3, and the
    parameter sensitivity analysis of Section 4.5 and Table 8. Both were
    conducted separately from the primary Study 3 evaluation in
    heldout_final_root36000000/, which remains the basis for the statistical
    claims of the article. Nothing here replaces it, and no parameter of the
    method was re-optimized as a result of it.
      representative_case/   selection protocol frozen before plotting, the
                             numerical selection record, the digest-verified
                             replay record, the analysis-only observer diff,
                             the figure, the plotted time series, and scripts
      PARAMETER_TABLE_DRAFT.md
                             the source-derived parameter inventory behind
                             Table 4, with calibration-status labels
      sensitivity/           protocol frozen before execution, task manifest
                             and hashes, all 720 run records, analysis script,
                             machine-readable results, compact table, and
                             completion record
      README.md              what each artifact is, with the reported values
      SHA256SUMS.txt         checksum of every file in that directory
    Everything else in this archive is unchanged from the originally submitted
    supplementary material.

SHA256SUMS.txt
    Checksum of every file in this archive.


INTEGRITY

Each result packet carries a "packet_sha256" field computed over its own
canonical content, so any modification is detectable. To verify the archive:

    sha256sum -c SHA256SUMS.txt

To regenerate the tables from the packets:

    python3 code/build_final_tables.py

That script re-verifies every packet checksum as it runs.


ENVIRONMENT

Results were produced with Python 3.12.3, NumPy 1.26.4 and OpenCV 4.6.0. The
optical front end requires an OpenCV build providing cv2.AKAZE_create.


LICENCE

Released under the MIT Licence together with the simulation software at
https://github.com/Irlkidonu/AUV-Simulator
