An efficient LiDAR-based semantic SLAM system that builds 3D semantic maps from laser scans.
SuMa++ is an open-source LiDAR-based semantic SLAM system that builds dense, semantically annotated 3D maps from laser range scans. It combines geometric mapping with deep learning-based semantic segmentation to enable robots to understand their environment beyond just geometry, recognizing objects like cars and buildings. The system is designed for real-time performance and is compatible with the KITTI dataset format.
Robotics researchers and engineers working on autonomous navigation, 3D mapping, and semantic perception using LiDAR sensors. It is particularly relevant for those developing self-driving cars, drones, or mobile robots that require environment understanding.
Developers choose SuMa++ for its efficient integration of semantic segmentation into a proven LiDAR SLAM pipeline, offering real-time semantic mapping without heavy computational overhead. Its compatibility with KITTI and support for GPU acceleration via TensorRT make it a practical tool for prototyping and research in semantic robotics.
SuMa++: Efficient LiDAR-based Semantic SLAM (Chen et al IROS 2019)
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Integrates RangeNet++ for online per-point semantic labeling during SLAM, enabling live environment interpretation as shown in the demo GIF.
Works directly with the standard KITTI Visual Odometry format, simplifying integration with widely used autonomous driving datasets and benchmarks.
Leverages TensorRT via rangenet_lib to speed up semantic segmentation, crucial for maintaining real-time performance with dense LiDAR scans.
Includes a Qt-based visualizer for monitoring the mapping process in real-time, aiding in debugging and result analysis, as mentioned in the README.
The README explicitly states it only works with KITTI dataset and semantic segmentation may not generalize to other environments without retraining or fine-tuning.
Requires building rangenet_lib separately, managing specific dependency versions like gtsam 4.0.0-alpha2, and Docker configuration for GUI visualization, increasing initial overhead.
Depends on NVIDIA GPUs for TensorRT and high OpenGL versions (≥4.0) for visualization, restricting use on systems without compatible graphics hardware.