A Dataset of Orthogonal Polygon-Derived Maze Environments for Path Planning Benchmarking

Version 1.3: This version includes updated benchmark results. This is the recommended version for use.

The dataset is organized into a structured directory format to facilitate access and reuse. It consists of orthogonal polygon data, corresponding grid-based maze representations, visualization images, and metadata files.

The primary purpose of this dataset is to provide a systematic collection of orthogonal polygon-derived maze environments for benchmarking path planning algorithms. By covering a wide range of geometric complexities and providing standardized start–goal configurations, the dataset supports consistent and reproducible evaluation across different algorithmic approaches.

Beyond benchmarking, the dataset can also be used in related application contexts. The grid-based maze representations are suitable for developing and testing navigation algorithms in artificial intelligence and robotics. In addition, the dataset can be applied in game development, particularly for procedural maze generation and navigation scenarios.

Furthermore, the visualization results in the images/path_visuals directory provide insight into the behavior of different path planning algorithms, including explored regions and final trajectories. These visual outputs may serve as data for training or evaluating learning-based approaches that aim to model or approximate path planning strategies.

The dataset covers a wide range of geometric complexities by considering 20 different vertex counts, ranging from 50 to 100,000 vertices. For each vertex count, 50 polygon instances are generated, resulting in a total of 1,000 orthogonal polygons and 1,000 corresponding maze instances.

The dataset/polygons/ directory contains orthogonal polygon instances grouped by the number of vertices. Each file stores a single polygon in CSV format, where each row corresponds to a vertex represented by its Cartesian coordinates (x, y). The vertices are ordered along the boundary of the polygon, forming a simple orthogonal shape without self-intersections.

The dataset/mazes directory contains grid-based representations derived from the corresponding polygon instances. Each maze is stored in JSON format and includes the grid size, a binary occupancy grid, and start–goal configurations. In the grid representation, cells are encoded as binary values, where 0 denotes free space and 1 denotes obstacles. The polygon boundary and exterior regions are treated as obstacles, while the interior defines the navigable space. The start and goal positions are generated using a two-sweep Breadth-First Search procedure, which approximates a long shortest path within the maze.

The dataset/images/mazes directory provides visualizations of the grid-based maze instances. This directory contains 1,000 images, each corresponding to a maze environment with annotated start and goal positions.

The dataset/images/path_visual/ directory contains visualization results of path planning algorithms applied to the maze instances. The subdirectory dataset/images/path_visual/six_algorithms contains 6,000 images illustrating the computed paths and, where applicable, the intermediate exploration structures of six algorithms: BFS, Dijkstra, A*, PRM, RRT-Connect, and RRT*. The images/path_visuals/rrt_star_optimized directory contains supplementary visualizations of the optimized RRT* algorithm. Under a specific algorithmic setup applied to the dataset in datasets/mazes/, this directory includes 502 PNG files. These images are provided only for instances where a valid path is successfully found during the extended multi-trial evaluation.

The metadata directory contains summary files that enable indexing and basic analysis. The file dataset_summary.csv provides an overview of all instances, including identifiers, number of vertices, file references, and start–goal configurations. The file pathfinding_summary.csv records benchmarking results for all algorithm–maze pairs, including success status, path length, and execution time.

Overall, the dataset comprises 1,000 polygon files (CSV), 1,000 maze files (JSON), 1,000 maze visualization images, and 6,502 path planning result images, providing a comprehensive resource for evaluating path planning algorithms across environments of varying complexity.

All data files are organized to allow direct use in computational experiments without additional preprocessing.

This dataset is associated with the data article:"A Dataset of Orthogonal Polygon-Derived Maze Environments for Path Planning Benchmarking" by Nguyen Kieu Linh.