Determining waypoints in neuroevolutionary ship control

In this article, the author presents a decision support system for a helmsman during ship maneuvering in a restricted area. The system is based on evolutionary artificial neural networks, whose structure and parameters, appropriately encoded in genetic code, evolve during simulation, adapting the entire population to the designated task of safely navigating the ship from a starting position to a destination. To improve the neural network learning process, the author proposes an algorithm for determining waypoints, which serve as an additional useful set of information when calculating rudder deflection.Neuroevolutionary ship control is a decision support system for a helmsman during ship maneuvering in a restricted area. In the simulated model, the ship is controlled by an artificial neural network. Its input consists of signals encoded from information regarding the current state of the environment in which the ship and the helmsman are located, e.g., current heading, speed over ground, and angular velocity. The network output provides rudder deflection angles.
In the evolutionary aspect, such a neural network is treated as an individual in a population of networks competing for the title of the best-adapted network for the studied problem [1], i.e., safely controlling the ship in a given restricted water area. Through appropriate selection while observing the current performance of helmsmen, the system selects the best helmsman and creates new individuals in the next generation through genetic operations. (…)

Determining waypoints in neuroevolutionary ship control – PDF file for download