Learned Uncertainty Calibration for Visual Inertial Localization
This addresses the issue of unreliable uncertainty estimates in visual inertial localization systems, which is incremental as it builds on existing EKF methods.
The paper tackled the problem of inaccurate covariance estimates in Extended Kalman Filters for visual inertial localization, showing that these inaccuracies are systematic and can be corrected by learning a nonlinear map from empirical ground truth to estimated covariance, as demonstrated on both simulated and real-world data.
The widely-used Extended Kalman Filter (EKF) provides a straightforward recipe to estimate the mean and covariance of the state given all past measurements in a causal and recursive fashion. For a wide variety of applications, the EKF is known to produce accurate estimates of the mean and typically inaccurate estimates of the covariance. For applications in visual inertial localization, we show that inaccuracies in the covariance estimates are \emph{systematic}, i.e. it is possible to learn a nonlinear map from the empirical ground truth to the estimated one. This is demonstrated on both a standard EKF in simulation and a Visual Inertial Odometry system on real-world data.