GUMin

GUMin is a software library and toolkit designed for efficient uncertainty minimization in large-scale estimation problems. It provides algorithms and data structures for building and solving factor-graph formulations of nonlinear least-squares problems, with emphasis on incremental updates, robust estimation, and performance on resource-constrained systems.

Originally developed from academic research into smoothing, mapping and sensor fusion, GUMin focuses on modularity and extensibility. Core components include a factor-graph representation, a suite of linearization and solver backends, support for robust loss functions, and utilities for marginalization and variable ordering. The library typically offers bindings in compiled and scripting languages to accommodate both embedded and rapid-prototyping workflows.

GUMin is applied in domains that require fusing heterogeneous measurements under uncertainty, such as robotics, simultaneous localization and mapping (SLAM), computer vision, and navigation. Typical workflows use GUMin to construct factors from sensor observations, optimize state estimates, and query marginal covariances or consistency metrics.

Architecturally, GUMin emphasizes a plugin-friendly design that enables swapping linear solvers, preconditioners, and backend numerical libraries. Interoperability features often include common serialization formats, integration helpers for middleware and visualization tools, and a test suite for reproducible experimentation. Community contributions and ongoing maintenance generally target scalability, numerical robustness, and ease of integration into larger systems.