Search Mechanisms in Geometry-Constrained Black-Box Optimization: A Benchmark Study on Continuous 3D UAV Path Planning

Overview This repository contains the complete dataset, data processing pipelines, and analytical scripts associated with the research paper:A Systematic Benchmark of Heuristic and Metaheuristic Optimization Algorithms for Continuous 3D UAV Path Planning. The purpose of this deposit is to ensure full transparency and complete reproducibility of the benchmarks, statistical analyses, and figures presented in the study. Dataset Description To overcome the storage and performance limitations of handling millions of raw individual CSV files generated during simulation runs, the data has been aggregated, compressed, and structured using Apache Parquet partitions. This approach preserves strict column data types while dramatically reducing storage size. The benchmark data inside data/results/ is divided into five specialized analytical datasets: summary: High-level performance metrics per individual optimization run. metadata: Simulation and algorithm-specific hyperparameters. history: Optimization history tracked across successive iterations. best_paths: Spatial coordinates of the best-generated continuous 3D paths. path_iters: Intermediate path states during the optimization process (large-scale dataset). Each of these directories is partitioned by the terrain complexity attribute: terrain=light terrain=moderate terrain=dense terrain=cluttered Additionally, the checkpoints directory contains native JSON files representing the tracked state of each unique simulation environment configuration. Repository Content & Workflow The uploaded archive contains the following structure: data/results/: The complete partitioned Parquet and JSON dataset described above. notebooks/UAV_pp_benchmarks_integrated.ipynb: The core simulation engine used to run the continuous 3D path planning benchmarks. notebooks/benchmark_pipeline.ipynb: The orchestration pipeline that reads raw outputs and compresses them into the final Parquet layout (utilizes src/pipeline_worker.py). notebooks/analysis.ipynb: The main analysis script containing the data loader, statistical tests, and chart generation code. How to Reproduce the Results To instantly reproduce the charts and statistics from the paper: Extract the archive and ensure Python 3.10+ is installed. Install dependencies: pip install pandas pyarrow fastparquet jupyter matplotlib seaborn Open and run notebooks/analysis.ipynb. This notebook includes a standalone data loader that reads the partitioned Parquet files directly into unified Pandas DataFrames, allowing you to recreate all findings immediately without re-running the heavy simulation backend. License Source Code: MIT License (included in the archive). Datasets & Results: Creative Commons Attribution 4.0 International (CC BY 4.0).