NHC-LIO: A Novel Vehicle Lidar-Inertial Odometry (LIO) With Reliable Nonholonomic Constraint (NHC) Factor

Current vehicle Lidar-inertial odometry (LIO) is prone to error accumulation, particularly when low-cost equipment is utilized in large-scale scenes without loop closure. In this article, we propose a novel approach to address this problem by introducing the ground nonholonomic constraint (NHC) and enhancing LIO with the NHC factor, referred to as NHC-LIO. The conventional NHC model assumes zero values for the lateral and vertical velocities, which proves to be challenging to adapt to complex and rapidly changing practical applications. Thus, we further design a convolutional neural network that applies a nine-axis inertial measurement unit (IMU) to predict reliable NHC pseudo-observation in real-time. During the factor graph optimization stage, the NHC noises and the corresponding information matrix are calculated adaptively based on the prior vehicle motion state, further enhancing the robustness of the LIO assisted with the NHC factor. Extensive evaluation and verification of the NHC-LIO using the KITTI dataset demonstrates its state-of-the-art performance. The NHC factor effectively mitigates LIO drift, especially in the vertical direction when loop closure detection is not available.