A Neural Network-based Unmanned Ship Autonomously Berthing Controller

The output of the neural network berthing controller is the form of engine/helm order instead of propeller revolution speed/rudder angle control commands as traditional controllers may give. Doing this is to emphasize the mapping between engine order/helm order and propeller speed/rudder angle which is ignored by traditional controllers. The training of the network is changed from multiple initial states into single initial state. In addition, the training inputs are extended to cover all information extracted from birthing processes while they are tailored in traditional controllers. Simulation is performed to verify the advantages of the controller.