MIT-SPARK/Kimera-RPGO

What it solves

Kimera-RPGO provides a robust way to perform Pose Graph Optimization (PGO), which is essential for robots to maintain an accurate map of their environment and their own position. It specifically addresses the problem of "outliers"—incorrect loop closure measurements that can lead to significant errors in a robot's estimated trajectory if not properly rejected.

How it works

The library acts as an optimization backend that uses a robust solver (defaulting to Levenberg-Marquardt or Gauss-Newton) to refine the robot's pose graph. It implements outlier rejection based on Pairwise Consistent Measurement Set Maximization (PCM), which identifies and removes inconsistent measurements to ensure only reliable data is used for the final optimization.

Who it’s for

It is designed for robotics developers and researchers working on SLAM (Simultaneous Localization and Mapping) and multi-robot map merging who need a reliable backend to handle noisy or incorrect sensor data.

Highlights

• Outlier Rejection: Uses PCM to filter out incorrect loop closures.
• Flexible Solvers: Supports both Levenberg-Marquardt (LM) and Gauss-Newton (GN) optimization methods.
• Dimensionality Support: Capable of handling both 2D and 3D pose graph optimization.
• Integration Ready: Can be used as a library integrated into a larger system or as a standalone tool for processing g2o files.