GTSAM (Georgia Tech Smoothing and Mapping)

A factor-graph optimization library from Georgia Tech, a common backend for SLAM and sensor fusion.

GTSAM is an open-source C++ library (BSD-licensed, with Python and MATLAB bindings) developed by Frank Dellaert's team at Georgia Tech. It represents an estimation problem as a factor graph: variables are the unknowns being estimated, such as poses and velocities, and factors are the constraints coming from sensor measurements, with nonlinear optimization then finding the most likely solution. Its distinguishing feature is the iSAM2 incremental solver, which, when a new measurement arrives, updates only the parts of the solution it affects, making it well suited to real-time SLAM; it also has built-in support for common factors like IMU preintegration. Lidar-inertial SLAM systems such as LIO-SAM use it as their back end. It's in the same category as g2o and Ceres Solver, though GTSAM leans more toward probabilistic modeling.

ExampleLIO-SAM feeds lidar odometry, IMU preintegration, GPS, and loop-closure constraints into GTSAM as factors, using iSAM2 to optimize the robot's trajectory in real time.

Related

Factor Graph Optimization · Simultaneous Localization and Mapping (SLAM) · IMU Preintegration · g2o (General Graph Optimization) · Ceres Solver · LIO-SAM

Sources

GTSAM 官网 (Chinese)
borglab/gtsam (GitHub)