This paper addresses the challenge of real-time visual-inertial odometry (VIO) by preintegrating inertial measurements between selected keyframes into single relative motion constraints. The authors develop a preintegration theory that properly handles the manifold structure of the rotation group, addressing issues such as rotation noise and singularities. They derive the maximum a posteriori (MAP) state estimator and analytic Jacobians for optimization and bias correction. The preintegrated IMU model is integrated into a factor graph framework, enabling the application of incremental smoothing algorithms and a structureless model for visual measurements, which avoids optimizing over 3D points. Extensive evaluations on real and simulated datasets demonstrate the effectiveness of the proposed approach, achieving accurate state estimation at 100 Hz, outperforming state-of-the-art methods. The paper also provides supplementary material, including video experiments and source code for the preintegrated IMU and structureless vision factors.This paper addresses the challenge of real-time visual-inertial odometry (VIO) by preintegrating inertial measurements between selected keyframes into single relative motion constraints. The authors develop a preintegration theory that properly handles the manifold structure of the rotation group, addressing issues such as rotation noise and singularities. They derive the maximum a posteriori (MAP) state estimator and analytic Jacobians for optimization and bias correction. The preintegrated IMU model is integrated into a factor graph framework, enabling the application of incremental smoothing algorithms and a structureless model for visual measurements, which avoids optimizing over 3D points. Extensive evaluations on real and simulated datasets demonstrate the effectiveness of the proposed approach, achieving accurate state estimation at 100 Hz, outperforming state-of-the-art methods. The paper also provides supplementary material, including video experiments and source code for the preintegrated IMU and structureless vision factors.