ecg2o: a seamless extension of g2o for equality-constrained factor graph optimization.

Area of Science:

• Robotics
• Computer Vision
• Optimization

Background:

• Factor graph optimization is crucial for robotic perception tasks like SLAM and SfM.
• Existing methods often use unconstrained least squares, limiting accuracy and applicability.
• Handling hard equality constraints in factor graphs is challenging, with prior work using soft penalties or complex Augmented Lagrangian methods.

Purpose of the Study:

• To develop a novel extension for factor graphs that seamlessly integrates hard equality constraints.
• To maintain the efficiency and flexibility of existing second-order optimization techniques while ensuring constraint satisfaction.
• To provide an open-source C++ library (ecg2o) for hard equality-constrained optimization in factor graphs.

Main Methods:

• Proposed a novel extension to factor graphs for native hard equality constraint incorporation.
• Implemented and benchmarked the method against Augmented Lagrangian baselines in g2o and GTSAM.
• Developed ecg2o, a header-only C++ library extending g2o for equality-constrained optimization.

Main Results:

• The novel approach successfully incorporated hard equality constraints without additional optimization layers.
• Demonstrated improved state estimation accuracy and broader applicability in an autonomous vehicle optimal control problem.
• Validated the method's efficiency and flexibility compared to existing techniques.

Conclusions:

• The proposed method offers an efficient and flexible way to handle hard equality constraints in factor graph optimization.
• This advancement can lead to more accurate state estimates and expanded applications in robotics and control.
• The open-source ecg2o library facilitates the adoption of these enhanced factor graph optimization techniques.