Abstract
This paper presents a method for estimating the velocity of underwater vehicles using LSTM-based neural networks and non-ultrasonic sensor data. While previous studies have focused on Autonomous Underwater Vehicles (AUVs), this study considers the distinct shape and motion characteristics of Remotely Operated Vehicles (ROVs) to effectively utilize surface pressure data. Data was collected by operating an ROV equipped with external pressure sensors and a Doppler Velocity Log. Input data included surface pressure, Inertial Measurement Unit accelerations and orientations, Remote Control inputs, and motor output, all reflecting the dynamic state of the ROV. By employing repeated movement macros, the dataset captured consistent motion patterns, allowing for robust training results against sensor measurement disturbances with minimal data adjustments. Additionally, the contribution of each sensor input to velocity prediction was evaluated to provide insight into their individual impacts. The proposed method demonstrated reliable velocity estimation in environments without external currents, overcoming the limitations of conventional ROV approaches reliant on ultrasonic sensors.
| Original language | Korean |
|---|---|
| Pages (from-to) | 252-259 |
| Number of pages | 8 |
| Journal | 로봇학회 논문지 |
| Volume | 20 |
| Issue number | 2 |
| DOIs | |
| State | Published - Jun 2025 |
Keywords
- Remotely Operated Vehicle
- Non-Acoustic Sensor
- Velocity Estimation
- LSTM
Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver