MCLIVO: A low-drift LiDAR-inertial-visual odometry with multi-constrained optimization for planetary mapping

In planetary exploration and mapping, robust and accurate state estimation is a critical functionality supporting the navigation and terrain construction of exploration rovers. Due to the lack of sufficient features in a planetary environment, leading to less constraints in drift case and limited robustness in localization. To address this situation, we propose a low-drift light detection and ranging (LiDAR)-inertial-visual odometry with multi-constraint optimization system. The proposed constraints including LiDAR-inertial constraint, visual-inertial constraint and ground manifold constraint are employed to perform optimization on the factor graph. We realize tightly-coupled LiDAR-inertial odometry designed by an iterative error-state Kalman filter (ESKF), and propose the rotational residual and preintegrated velocity residual besides the point-to-plane residual to improve the estimation accuracy of state variables. In the visual-inertial odometry, we propose a special visual-LiDAR association strategy that considers planetary terrain to improve the reliability of feature depth estimation. To limit the drift in the vertical Z-axis direction, we propose a ground manifold estimation constraint composed of distance residual and coplanar residual. Furthermore, we propose a two-stage sensor degenerate observer that enables the system more resilient and reliable in the featureless planetary environment. Real-world experiments in imitative planetary environments have shown that our method has the higher localization precision and robustness compared to other excellent sensor coupling schemes.