SpatialLM
Let AI "see" 3D space — point clouds to 3D bounding boxes/floor plans/IFC
Lets AI "see" three-dimensional space: point clouds converted to 3D bounding boxes, floor plans, and IFC.
Robotics and spatial AI — machines that see and move.
Key features
- 3D large language model from point clouds to structured scene output
- Detects walls, doors, windows, oriented object boxes with categories
- Accepts point clouds from monocular video, RGBD images, LiDAR
- User-specified category detection in v1.1; NeurIPS 2025
Good for
- 3D indoor scene reconstruction
- Embodied robotics and autonomous navigation
Stack & tags
PythonPyTorch3dpoint-cloudscene-understandingobject-detectionopen-weightsrobotics3d-reconstruction
Quick start
python inference.py --point_cloud pcd/scene0000_00.ply --output scene0000_00.txt --model_path manycore-research/SpatialLM1.1-Qwen-0.5BFrom the project README on GitHub.
Similar open-source projects
NVIDIA GR00T N1
★ 8,150The world's first open-source generalist humanoid robot foundation model
OpenGlass
★ 4,175Turn any glasses into AI-powered smart glasses