G-solver: Gaussian Belief Propagation and Gaussian Processes for Continuous-Time SLAM

Davide Ceriola, Simone Ferrari, Luca Di Giammarino, Leonardo Brizi, Giorgio Grisetti

Abstract: Continuous-time Simultaneous Localization and Mapping (CT-SLAM) offers a natural way to integrate data from diverse and asynchronous sensors. Most existing approaches rely on centralized Nonlinear Least Squares (NLLS) solvers, which are hard to scale. In contrast, distributed Gaussian Belief Propagation (GBP) offers a scalable, decentralized approach that naturally manages uncertainty. However, existing GBP methods for continuous SLAM rely on spline-based interpolation, which is hard to tune and do not offers easy uncertainty extraction. Gaussian Processes (GPs) provide a robust alternative by modeling dynamics and their uncertainty. In this paper, we introduce a distributed GBP solver that uses GP priors for continuous-time trajectory estimation, resulting in improved accuracy and efficiency without compromising execution times, fundamental for real-world applications. We release an open-source implementation at TBA.

Submission Number: 10