A Non-Equal Time Interval Incremental Motion Prediction Method for Maritime Autonomous Surface Ships.

Area of Science:

• Maritime technology
• Robotics
• Data fusion

Background:

• Autonomous maritime surface ships rely on accurate sensor data for safe navigation.
• Disparate sensor sample rates can degrade the accuracy and reliability of fused data.
• Synchronizing data from sensors with different sampling frequencies is crucial for precise motion prediction.

Purpose of the Study:

• To propose a non-equal time interval incremental prediction method for maritime ship motion.
• To enhance the quality of fused perceptual data by accounting for varying sensor sample rates.
• To improve the precise anticipation of ship motion status at each sensor's sampling time.

Main Methods:

• Utilized the cubature Kalman filter to estimate ship motion at equal intervals.
• Developed a ship motion state predictor using a long short-term memory (LSTM) network.
• Input to the LSTM network included the increment and time interval of historical estimations; output was the motion state increment.

Main Results:

• The proposed method demonstrated a significant reduction in prediction error, with the root-mean-square error coefficient decreased by approximately 78% on average.
• The technique effectively mitigates the impact of speed differences between training and testing datasets on prediction accuracy.
• Algorithm execution times were comparable to traditional LSTM prediction methods, meeting engineering requirements.

Conclusions:

• The non-equal time interval incremental prediction method accurately forecasts ship motion status despite asynchronous sensor data.
• This approach enhances the reliability and precision of data fusion for autonomous maritime navigation.
• The method offers a practical solution for improving the safety and efficiency of autonomous ships.