Navigation and path planning solutions for underwater autonomous vehicles

The rapid advancement of technology has increased interest in unmanned underwater systems. These systems are utilized in various fields such as exploration, observation, and underwater operations. However, the process of determining the location underwater is not feasible with the widely used GPS technology in air and land systems. In response to this need, unmanned underwater vehicles have been developed to determine their locations using acoustic methods or GPS references obtained from surface stations in conjunction with underwater sensors. Positioning solutions in open seas can be expensive, prompting the exploration of more practical alternatives within enclosed spaces. This study involves a two-stage process. In the first stage, the underwater vehicle is placed in a closed pool environment and scans its surroundings using a 360-degree scanning sonar. The acquired scanning data is classified using the K-means algorithm, and the vehicle's position within the pool is determined using a corner detection method. Subsequently, the data from the distance sensor on the vehicle is modeled using machine learning algorithms. In the modeling process, Random Forest algorithm yielded the best results among Linear Regression, KNN, Decision Tree, and Random Forest algorithms. This model allows the prediction of the distance traveled by the underwater vehicle during its movement.