RV-LIO: reinforced-vector-based LiDAR-inertial odometry for weak-feature environments

This paper studies the light detection and ranging (LiDAR)-inertial odometry (LIO) invalidation problem for unmanned aerial vehicles (UAVs) in weak-feature environments. A reinforced-vector-based LIO (RV-LIO) system is developed to tackle this challenge. First, valid localization is proposed as a criterion based on constraint vector distribution in weak-feature environments. Two boundary conditions are derived from the relation between vector distribution and state observability. Second, a reinforced vector module is designed to extract and reinforce special constraint vectors through iterative singular value decomposition and noise coefficient adjustment, respectively. Third, a repaired vector module is built to capture and correct certain constraints by incorporating prior knowledge of tunnel surroundings. Both modules serve as subsystems of the filter process in the LIO system. Finally, Gazebo simulation and real experiment results are provided to verify the effectiveness of the proposed system.