2019/09/24 by Kevin Doherty, Doherty, Kevin, David Baxter +5 · 3 citations
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #FOS: Computer and information sciences #Indoor and Outdoor Localization Technologies #Robotics (cs.RO) #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1909.11213
openalex publication_date 2019/09/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Modern robotic systems sense the environment geometrically, through sensors\nlike cameras, lidar, and sonar, as well as semantically, often through visual\nmodels learned from data, such as object detectors. We aim to develop robots\nthat can use all of these sources of information for reliable navigation, but\neach is corrupted by noise. Rather than assume that object detection will\neventually achieve near perfect performance across the lifetime of a robot, in\nthis work we represent and cope with the semantic and geometric uncertainty\ninherent in methods like object detection. Specifically, we model data\nassociation ambiguity, which is typically non-Gaussian, in a way that is\namenable to solution within the common nonlinear Gaussian formulation of\nsimultaneous localization and mapping (SLAM). We do so by eliminating data\nassociation variables from the inference process through max-marginalization,\npreserving standard Gaussian posterior assumptions. The result is a\nmax-mixture-type model that accounts for multiple data association hypotheses\nas well as incorrect loop closures. We provide experimental results on indoor\nand outdoor semantic navigation tasks with noisy odometry and object detection\nand find that the ability of the proposed approach to represent multiple\nhypotheses, including the "null" hypothesis, gives substantial robustness\nadvantages in comparison to alternative semantic SLAM approaches.\n