Research on error compensation and trajectory optimization of underwater inertial odometer based on improved LSTM network

Autonomous Underwater Vehicles (AUVs) face numerous navigation challenges in complex underwater environments, where traditional navigation techniques such as sonar, visual sensors, and Inertial Navigation Systems (INS) have limitations. Underwater Inertial Odometry (UIM) estimates AUV motion trajectories by fusing data from Inertial Measurement Units (IMU) and odometry, but traditional methods lack robustness under complex underwater noise interference. This paper proposes an improved Long Short-Term Memory (LSTM) network structure that incorporates attention mechanisms and a multimodal fusion module. By leveraging bidirectional LSTM and residual connections, the model enhances training efficiency and long-term dependency modeling capabilities. Additionally, an error compensation and trajectory optimization model is constructed, incorporating underwater motion physical constraints through a joint loss function, with further trajectory smoothing optimization applied during post-processing. Experimental results demonstrate that this method outperforms traditional approaches and deep learning baselines on both public underwater datasets and self-collected data, validating the effectiveness of the proposed improvements in error compensation and trajectory optimization.