Neural Network-based Genetic Algorithm for Autonomous Boat Pathfinding

Genetic algorithms become widely used in various optimization as a nature-inspired algorithm. This biological-based algorithm includes three genetic operators: selection, crossover or recombination, and mutation. In this study, the genetic operator processed neural network parameters, including the weight and bias. Three-ray sensors with different directions input data into the neural network. We applied the method to autonomous boats for pathfinding responding to dynamic environment challenges. Generally, autonomous boats face difficulty in path-finding in dynamic environments. To evaluate the method, we experimented with the scenario of the static and dynamic environment (with wave surface) in a three-dimensional simulation platform. The hyperparameter of layer number and mutation rate variation were used. The results showed that the method was useful in pathfinding in static and dynamic environments.