A Decoupled and Linear Framework for Global Outlier Rejection over Planar Pose Graph

TL;DR

This paper introduces DEGNC-LAF, a fast and robust framework for planar pose graph optimization that effectively rejects outliers using a decoupled, linear, and non-convex approach, outperforming existing methods in speed and accuracy.

Contribution

It presents a novel decoupled, linear angle-based framework for outlier rejection in planar pose graph optimization, enabling efficient solutions without initial guesses.

Findings

Runs up to 30 times faster than standard GNC.

Achieves high-quality pose estimates despite outliers.

Validated extensively on benchmark datasets.

Abstract

We propose a robust framework for the planar pose graph optimization contaminated by loop closure outliers. Our framework rejects outliers by first decoupling the robust PGO problem wrapped by a Truncated Least Squares kernel into two subproblems. Then, the framework introduces a linear angle representation to rewrite the first subproblem that is originally formulated with rotation matrices. The framework is configured with the Graduated Non-Convexity (GNC) algorithm to solve the two non-convex subproblems in succession without initial guesses. Thanks to the linearity properties of both the subproblems, our framework requires only linear solvers to optimally solve the optimization problems encountered in GNC. We extensively validate the proposed framework, named DEGNC-LAF (DEcoupled Graduated Non-Convexity with Linear Angle Formulation) in planar PGO benchmarks. It turns out that it runs significantly (sometimes up to over 30 times) faster than the standard and general-purpose GNC while resulting in high-quality estimates.