ASSURE-BVLOS: Reference Implementation and Reproducibility Package for Runtime Assurance in Risk-Adaptive Space–Air–Ground Unmanned Aircraft Operations
Abstract
ASSURE-BVLOS is the executable reference implementation and reproducibility package supporting the revised manuscript: “ASSURE-BVLOS: Runtime Assurance for Risk-Adaptive Space–Air–Ground Unmanned Aircraft Operations.” The framework integrates five monitored risk domains—command-and-control integrity, sensing reliability, external-data integrity, environmental exposure, and vehicle health—with explicit uncertainty penalties, deterministic hard-limit overrides, supervisory authority states, and configuration-linked evidence. VERSION 2.0.0-REVISION This version contains a newly implemented and fully specified computational evaluation of the ASSURE-BVLOS framework. Important provenance notice: the executable implementation and state-level outputs underlying the earlier numerical evaluation were unavailable for verification. Version 2.0.0-revision therefore newly implements the documented framework and replaces the earlier numerical evaluation. It is not a recovered or verified reproduction of the original simulation. Previously unreported parameters are explicitly identified, and no parameter was selected for the purpose of reproducing the earlier headline result. PRIMARY COMPUTATIONAL EVALUATION The primary evaluation contains 100,000 independent synthetic states generated using NumPy PCG64 with a fixed random seed. Under the declared synthetic generator, outcome-surrogate model, authority rules, and mitigation assumptions: • Static baseline realised simulated hazardous-outcome proportion: 10.739%• ASSURE-BVLOS realised simulated hazardous-outcome proportion: 8.083%• Paired difference: 2.656 percentage points• 95% paired-bootstrap interval: 2.557–2.757 percentage points• Critical-state recall: 40.20%• S2–S4 intervention burden: 9.768% The revised evaluation additionally includes threshold sensitivity, domain ablation, weight sensitivity, uncertainty-penalty sensitivity, surrogate and mitigation analyses, alternative fusion comparisons, repeated random-seed evaluations, and prescribed-score temporal supervisory-logic tests.
Interpretation
The nominal direction of the simulated outcome difference is structurally favoured by the assumed mitigation schedule. These results therefore demonstrate reproducible behaviour of the configured supervisory model and should not be interpreted as validated real-world accident reduction, causal intervention effectiveness, operational flight-safety performance, or certification evidence. SUPPLEMENTARY CODE S1 — ESM_1.zip Contains: • executable Python reference implementation;• complete configuration and fixed random seeds;• all 100,000 primary synthetic states and paired outcomes;• generated numerical results;• threshold, weight, ablation, uncertainty, surrogate, mitigation, and repeated-seed analyses;• implementation tests and replay checks;• numerical and conceptual figure sources;• equation sources and audit information. SUPPLEMENTARY
Methods
S2 — ESM_2.pdf Contains: • parameter provenance;• complete synthetic generator specification;• normalisation and supervisory-authority algorithms;• statistical and paired-bootstrap methodology;• sensitivity-analysis definitions;• supplementary figures;• reproduction instructions and software-environment details. HIGH-RESOLUTION FIGURES ASSURE_BVLOS_High_Resolution_Figures.zip contains publication-quality versions of the revised main and supplementary figures. SOFTWARE AND REPRODUCIBILITY The reference execution used Python 3.13.5 with NumPy 2.3.5, SciPy 1.17.0, pandas 2.2.3, and Matplotlib 3.10.8. The supplied implementation passed 28 implementation tests and 65 output-consistency checks. Same-environment replay reproduced all 29 primary arrays and all 16 result CSV files used in the final computational package. These checks establish implementation-level reproducibility within the specified computational environment. They do not constitute formal software qualification, operational aircraft validation, or certification evidence. No UAS hardware experiment, hardware-in-the-loop validation, human-in-the-loop study, or controlled flight validation is claimed in this version. GitHub repository:https://github.com/Nick-Barua/ASSURE-BVLOS Earlier Version 1.0.0:https://doi.org/10.5281/zenodo.21642250
