State Estimation for Autonomous Surface Vehicles: A Neural Network-Assisted Observer Approach

Autonomous Surface Vehicles (ASVs) have emerged as a revolutionary solution to enhance search and rescue (SAR) operations in open waters. In order to successfully do SAR missions, the ability of the boat to autonomously perform exploration with accurate movements is crucial. This paper presents a method to accurately track a boat's movement only based on the boat's kinematic equation. To implement this method, a neural-network is implemented by training it with the real robot's movement coordinate. In this research, a boat prototype is designed and to track the boat's movement, a vision-based tracking method is implemented by using a drone. A video of the boat's movement is captured from above, then the robot's moving coordinate is extracted by utilizing OpenCV object tracking. By integrating the boat's kinematic equation with a trained neural-network to compensate for the dynamic uncertainties present in the physical world, the robot's movement can be accurately tracked even without the help of any sensor. There are a lot of uncertainties that present in the water, thus large set of data is needed to further improve the robot's performance. This paper serves as a preliminary study on whether a neural-network could be utilized for movement tracking of a boat. To further improve the robot's movement tracking ability, more sensors could be utilized with added control system for position tracking ability.