Handling Local and Global Ambiguities via a Generalized Graph SLAM Framework based on Multimodal and Hyperedge Constraints

Graph-based Simultaneous Localization and Mapping (SLAM) has experienced a recent surge towards robust methods. These methods take the combinatorial aspect of data association into account by allowing decisions of the graph topology to be made during optimization. In this paper, the Generalized Graph SLAM framework for SLAM under ambiguous data association is presented, and a formal description of using hyperedges to encode uncertain loop closures is given for the first time. The framework combines both hyperedges and multimodal Mixture of Gaussian constraints to cover all sources of ambiguity in SLAM. An extension of the authors’ multimodal Prefilter method is developed to find good initial conditions in such a generalized multimodal hypergraph. Experiments on synthetic datasets show that the novel multimodal hypergraph Prefilter method is both significantly more robust and faster than other robust state-of-the-art methods.