Attention-Augmented LSTM Feed-Forward Compensation for Lever-Arm-Induced Velocity Errors in Transfer Alignment

Area of Science:

• Robotics and Control Systems
• Navigation and Guidance
• Bio-inspired Systems

Background:

• Underwater robotic systems face navigation challenges due to time-varying lever arms caused by structural flexure and attitude changes.
• Traditional rigid body compensation methods fail with flexible components, leading to velocity errors and degraded transfer alignment accuracy.
• Accurate navigation is crucial for the successful deployment and operation of mother-child underwater robotic systems.

Purpose of the Study:

• To develop a robust transfer alignment method for underwater robotic systems with time-varying lever arms.
• To enhance the accuracy and filter convergence of inertial navigation systems (INSs) in flexible robotic platforms.
• To mitigate systematic velocity errors introduced by structural deformations and attitude variations.

Main Methods:

• Augmenting a velocity-attitude joint matching and innovation-based adaptive Kalman filter (AKF) with an attention-based Long Short-Term Memory (LSTM) feed-forward module.
• Utilizing a short, real-time Inertial Measurement Unit (IMU) sequence from the slave INS to predict and compensate velocity bias.
• Implementing numerical simulations for an underwater bio-inspired robot deployment scenario to validate the proposed method.

Main Results:

• The proposed method significantly reduced the root-mean-square (RMS) misalignment angle error from approximately 14.5' to 5.2'.
• RMS installation error angle was reduced from 8.8' to 3.0', representing average reductions of about 64% and 66%, respectively.
• Demonstrated substantial improvement in the robustness and practical applicability of transfer alignment under dynamic conditions.

Conclusions:

• The LSTM-augmented AKF framework effectively compensates for velocity bias induced by time-varying lever arms in flexible underwater robotic systems.
• The enhanced transfer alignment method improves navigation accuracy and filter convergence, outperforming traditional rigid body compensation.
• The approach offers improved robustness and practical applicability for complex underwater robotic missions.