koide3/glim

GLIM (Global Localization and Inertial Mapping) represents a high-water mark for open-source LiDAR SLAM frameworks. With over 1,500 stars and significant forks, it is a successor to the widely used hdl_graph_slam. Its primary moat is its modularity; unlike specialized frameworks like Fast-LIO2 (which focuses on speed via IEKF) or LIO-SAM (which is robust but rigid), GLIM's plugin-based architecture allows developers to swap out estimation backends, deskewing methods, and loop closure detection. This makes it highly attractive for commercial robotics companies that need to adapt to different hardware configurations (e.g., Ouster vs. Livox vs. Velodyne). The 4-year project age and the author's reputation in the field provide significant community lock-in. While 'Frontier Labs' like OpenAI are unlikely to build a niche LiDAR framework, the long-term threat comes from the shift toward 'Neural SLAM' (using Gaussian Splatting or NeRFs for localization), though traditional geometric SLAM remains the production standard for reliability and low latency on edge hardware. Platform risk is medium as NVIDIA (through Isaac SDK) could provide more optimized, hardware-specific alternatives that erode the need for generic frameworks.