Abstract
This paper presents normalizing flows for incremental smoothing and mapping (NF-iSAM), a novel algorithm for approximating the full posterior distribution in SLAM problems with nonlinear measurement models and non-Gaussian factors. NF-iSAM exploits the expressive power of neural networks, and trains normalizing flows to model and sample the full posterior. By leveraging the Bayes tree, NF-iSAM enables efficient incremental updates similar to iSAM2, but in the more challenging non-Gaussian setting. We demonstrate the advantages of NF-iSAM over state-of-the-art point- and- distribution estimation algorithms using range-only SLAM problems with data association ambiguity. Our experimental evaluations show that NF-iSAM provides superior accuracy in describing the posterior beliefs of both continuous variables (e.g. position) and discrete variables (e.g. data association).