VDBFusion: Flexible and Efficient TSDF Integration of Range Sensor Data.

Area of Science:

• Robotics
• Computer Vision
• 3D Reconstruction

Background:

• Robotic mapping is essential for real-world navigation.
• Environment representation varies based on robot tasks and sensors.
• Existing methods often have limitations regarding environment size or sensor types.

Purpose of the Study:

• To develop a practical and efficient volumetric surface reconstruction system.
• To create a flexible mapping approach adaptable to various sensors and environments.
• To provide an accessible library for volumetric mapping challenges.

Main Methods:

• Utilized truncated signed distance functions (TSDFs) for 3D map representation.
• Leveraged the OpenVDB library for effective volumetric data handling.
• Integrated point cloud data from a 64-beam LiDAR sensor.

Main Results:

• Achieved real-time integration of LiDAR point clouds at 20 FPS on a single CPU core.
• Demonstrated a flexible system capable of multi-domain and multi-sensor applications.
• Developed an efficient and effective 3D map representation.

Conclusions:

• The proposed TSDF-based system offers a robust and efficient solution for real-world robotic mapping.
• The system's flexibility and performance make it suitable for diverse robotic applications.
• An accompanying C++ and Python library facilitates easy implementation of volumetric mapping.