Forest	RMUC Field
1.0 M Points	2.3M Points
0.45 seconds	0.62 seconds

• Overview
• Key Features
• Dependencies
• Supported Versions
• Installation
• Quick Start
• How to Use
• Visualization
• Core Concepts
• Performance
• Contributing
• Thanks
• License

This project is a Unity package for real-time point cloud generation based on scene colliders. It provides two primary methods:

• VoxelGeneratorOverlap: Utilizes an octree-like approach, performing parallel overlap detections from coarse to fine levels. Ideal for generating spatial voxel point clouds.
• VoxelGeneratorRaycast: Casts rays from the sky downwards to detect the ground. Best suited for generating terrain or ground-level point clouds.

Both methods leverage Unity's C# Job System, Burst Compiler, and Native Collections for high-performance, parallel computation.

• Real-time Generation: Generate point clouds for large-scale scenes in real-time.
• High Performance: Utilizes a hierarchical parallel detection strategy for efficient voxel generation.
• Data Export: Supports exporting point cloud data to .ply and .pcd files.
• Extensible: Designed with a clean and modular code structure for easy customization and extension.

This package relies on the following Unity packages. Please ensure they are installed in your project:

• com.unity.jobs
• com.unity.burst
• com.unity.collections

Unity 2022.3 or newer

• Note: Unity 2023 is not recommended due to instability in the core OverlapSphereCommand and RaycastCommand APIs.

You can install this package via Git URL in Unity Package Manager:

https://github.com/TJU-Aerial-Robotics/VoxelGenerator.git?path=/Packages/VoxelGenerator

• Navigate to Window > Package Manager.
• Click on the + icon in the top left corner.
• Select Add package from git URL....
• Paste the URL above and click Add.

1. Create a project with Unity 2022.3 or a newer version.
2. Install the package via Unity Package Manager using the Git URL provided above.
3. Import the Sample Scene from the package to your project.
   • Navigate to Window > Package Manager.
   • Select the VoxelGenerator package and click on Import Sample.
   • Choose the SampleScene to import.
4. Open the SampleScene and enter Play mode to see the point cloud generation in action.

1. Add either the VoxelGeneratorOverlap or VoxelGeneratorRaycast prefab to your scene.
2. Adjust the parameters in the Inspector to define the generation area and precision.
3. Use the StartGeneration() and ClearPointCloud() public methods to control the generation process from your scripts. Call the GetVoxelPointCloudData() public method to retrieve data. Alternatively, use the built-in shortcuts: press P to start generation and R to clear the point cloud.

Key Parameters (VoxelGeneratorBase):

• Voxel Size: The size of the smallest voxel.
• Max Depth: The maximum depth of the octree subdivision.
• Target Layer: The physics layer to detect colliders on.
• File Name: The name of the output .ply or pcd file.

You can use the Pcx package, which provides a simple way to render point clouds in Unity, or use external tools like CloudCompare or Meshlab to visualize the generated point clouds.

This method recursively subdivides a defined bounding box into eight octants (an octree). At each level of depth, it performs a batch of OverlapSphereCommand checks in parallel to determine if any colliders are present within each octant. If an octant is occupied and the max depth has not been reached, it is subdivided further. This continues until the finest voxel level is reached, generating a dense spatial point cloud.

This method creates a grid of points above the target area and casts rays downwards using RaycastCommand. The hit points on the collider surfaces form the resulting point cloud. It is highly efficient for creating 2.5D height maps or ground scans.

Both methods are optimized for performance using Unity's Job System and Burst Compiler. The VoxelGeneratorOverlap method is particularly efficient for large scenes with complex geometry, while the VoxelGeneratorRaycast method excels in scenarios where terrain or ground-level data is required.

Snapshot
Total Points	0.66 M	2.19 M	5.15 M	10.02 M
Flightmare	7.01s	20.47s	45.23s	85.31s
VoxelGenerator	0.35s	0.45s	0.82s	0.99s

Snapshot
Total Points	0.25 M	0.56 M	1.00 M	2.25M
VoxelGenerator	54ms	35ms	52ms	129ms

Contributions are welcome! Please feel free to submit a pull request. For more details, see CONTRIBUTING.md.

This project is inspired by Flightmare.

If you use this package in your research or projects, please cite it as follows:

@misc{YOPO-Rally,
      title={YOPO-Rally: A Sim-to-Real Single-Stage Planner for Off-Road Terrain}, 
      author={Hongyu Cao and Junjie Lu and Xuewei Zhang and Yulin Hui and Zhiyu Li and Bailing Tian},
      year={2025},
      eprint={2505.18714},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2505.18714}, 
}

This project is licensed under the Apache-2.0 License.