Analytic IMU Preintegration That Associates Uncertainty on Matrix Lie Groups for Consistent Visual-Inertial Navigation Systems

We present a novel right-invariant inertial measurement unit (RI IMU) preintegration model and apply it to an optimization-based visual-inertial navigation system (VINS). We find that the unobservable subspace of the proposed estimator was only affected by landmarks from the standpoint of an observability analysis. We highlight that the proposed VINS utilizing the RI IMU preintegration model is consistent with Jacobians evaluated using the most recent estimates of extended poses. Monte Carlo simulations and real-world experiments using the 4Seasons data sets validated our analysis and demonstrated the effectiveness of the proposed VINS by comparing its performance with VINS using different IMU preintegration models and other state-of-the-art VINS.