GitHub repository
A Siamese neural network for LiDAR-based loop closing and localization by predicting scan overlap and relative yaw angle.
What is GitHub repository?
OverlapNet is a deep learning system for 3D LiDAR-based SLAM that performs loop closing by predicting the overlap and relative yaw angle between LiDAR scans. It uses a modified Siamese network to process range images and other modalities like normals and intensity, enabling robots to recognize previously visited locations and correct accumulated odometry errors. The project originated from research at the University of Bonn and was nominated for Best System Paper at Robotics: Science and Systems (RSS) 2020.
Robotics researchers and engineers working on LiDAR-based SLAM, autonomous navigation, and place recognition, particularly those interested in integrating deep learning into geometric perception pipelines.
It provides a data-driven alternative to traditional loop closure methods, offering higher robustness in challenging environments by learning multi-modal scan similarities. The open-source implementation includes full training and inference pipelines, pre-trained models, and demos using real-world datasets like KITTI.
Overview
OverlapNet - Loop Closing for 3D LiDAR-based SLAM (chen2020rss)
Use Cases
Best For
- Adding deep learning-based loop closure to existing LiDAR SLAM systems
Related Projects
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
