Surrogate Models for MTSOS-Based UAV Flight-Time Prediction: Code and Data

This repository provides the code and numerical data accompanying a study of surrogate models for MTSOS-based flight-time prediction along fixed three-dimensional UAV paths. The study evaluates prediction accuracy, candidate-path ranking and computational cost under prescribed kinematic constraints. The main dataset comprises 18,000 cubic B-spline paths: 10,000 training paths and four test sets containing 2,000 paths each. The test domains cover in-distribution conditions and separate shifts in control-point count, path length and both properties simultaneously. Candidate paths are grouped into tasks containing 50 alternatives, with predefined task-level folds for cross-validation. Reference flight times are computed using MTSOS with a fixed spatial mesh of 0.05 m. Surrogate features are extracted using nominal spacings of 1, 2, 5, 10 and 20 m, together with an adaptive sampling variant. Feature sampling and the MTSOS reference mesh are distinct experimental settings. The repository contains: Python and native solver code for path generation, reference calculations, feature extraction, model training, prediction and timing; original and final path geometries, reference labels and replacement provenance; trained model artifacts, prepared features and per-path predictions; numerical accuracy, ranking, seed-stability and computational-time results; experimental configurations, random seeds, fold assignments and provenance metadata. The README documents the directory structure, software environment and execution procedures. Figures and detailed interpretation of the results are provided in the accompanying manuscript and supplementary material.