An optimization method for visual-inertial odometry enhanced by multiplehomogeneous IMUs

Visual-Inertial Navigation Systems (VINS) integrate visual information with inertial measurement unit (IMU) data, offering a cost-effective and practical solution for precise pose estimation in intelligent platforms such as unmanned aerial vehicle (UAV) and autonomous ground vehicle. However, conventional VINS suffers from significant drift due to IMU sensor errors and visual feature insufficiencies in certain environments. With advances in hardware manufacturing and the reduction in sensor cost and size, state estimation using multiple homogeneous sensors has become increasingly feasible. This paper proposes a multi-IMU visual-inertial odometry (VIO) framework to achieve low-drift continuous pose estimation. Specifically, we extend the classical multi-state constraint filter (MSCKF) by incorporating redundant IMU states and leveraging rigid-body rotational and translational constraints between multiple IMUs to suppress system drift effectively. Real-world vehicle experiments validate the proposed approach, demonstrating significant improvements over single-IMU VIO. Compared to the single-IMU case, our VIO framework reduces median position errors by 64% and 69% in the east (E) and north (N) directions, respectively. Velocity estimation accuracy improves by 66%, 63%, and 67% along the ENU axes, while the mean absolute yaw error decreases by 62%. Additionally, the incorporation of redundant IMUs significantly enhances the observability of gyroscope and accelerometer biases.