Provides SO3, SE3, and S2 manifold implementations with exponential/log maps, Jacobians, and composition operations. Self-contained (only depends on Eigen), well-tested for numerical stability, and directly reusable in any robotics project requiring Lie group algebra. Demonstrates high craftsmanship in handling singularities and small-angle approximations.

Implements an octree-based voxel map with a precomputed search order for efficient K-nearest-neighbor (KNN) queries. Uses a tsl::robin_map for hash-based voxel lookup and a custom KNN heap. The search order is optimized for early termination based on distance thresholds. This component is a core algorithmic contribution of the paper and can be extracted for use in other LiDAR SLAM or point cloud processing systems.

A thread-safe ring buffer with blocking and non-blocking pop operations, using std::mutex and std::condition_variable. Supports timeout-based waiting and move semantics. The design prioritizes real-time responsiveness over data completeness (old data may be overwritten). This is a clean, reusable concurrency primitive for sensor data buffering.