Probabilistic Appearance-Invariant Topometric Localization With New Place Awareness

Probabilistic state-estimation approaches offer a principled foundation for designing localization systems, because they naturally integrate sequences of imperfect motion and exteroceptive sensor data. Recently, probabilistic systems utilizing appearance-invariant visual place recognition (VPR) methods as the primary have demonstrated state-of-the-art performance in presence substantial appearance change. However, existing 1) do not fully utilize odometry data within models, 2) are unable to handle route deviations, due assumption that query traverses exactly repeat mapping traverse. To address these shortcomings, we present new topometric system which incorporates full 3-dof into model furthermore, adds an "off-map" state framework, allowing feature significant detours from reference map be successfully localized. We perform extensive evaluation on multiple Oxford RobotCar dataset exhibiting both change deviations routes previously traversed. In particular, evaluate two practically relevant tasks: loop closure detection global localization. Our approach achieves major improvements over improved systems.