An Efficient Trajectory Estimator Based on Probabilistic Inference

This article presents a probabilistic inference-based trajectory estimator (PITE) for motion estimation utilizing light detection and ranging (LiDAR) and inertial measurement unit (IMU) data. The approach introduces a split Gaussian process trajectory model, which decouples the rotational and translational trajectories to reduce computational complexity. An efficient closed-form inference method is proposed by leveraging the trajectory model's sparsity, enabling the marginalization of previous states and achieving globally optimal estimation. The fusion process and likelihoods are designed to integrate LiDAR and IMU measurements into the probabilistic trajectory model. Ablation studies validate the effectiveness of the components in the developed PITE. Comparative experiments across various scenarios demonstrate that PITE outperforms other existing algorithms. Furthermore, real-time flight experiments are conducted to validate the performance of the proposed strategy. Similar experiments have not been performed in previous trajectory estimation works.