AdaNGA: Adaptive neuronal genetic algorithms for unmanned surface vehicle dynamic path planning in dynamic environment

Dynamic path planning for unmanned surface vehicles (USVs) in dynamic water environments remains a significant challenge due to the dynamic changing obstacles and conditions. To address the challenges, this study introduces AdaNGA (an Adaptive Neuronal Genetic Algorithm) that enhances navigation by integrating neural networks with an adaptive genetic algorithm framework. The main novelty lies in a dynamic fitness function that changes based on obstacle type (static or dynamic), enabling real-time adaptation and efficient path optimization. This study implemented a 3D simulation using Unity 3D platform, incorporating real-time sensor inputs, real-world parameters (wavy water surface, water drag, air drag, buoyancy force, water current, and wind), physics parameters (inertia and thruster delay) and tested on the real-world map-based simulation with dynamic water environment (the Dammam Corniche beach area, Dammam City, Saudi Arabia). The proposed method AdaNGA demonstrates superior convergence speed and reduced travel time compared other basic metaheuristic algorithms (simulated annealing, particle swarm optimization, and basic genetic algorithms) and baseline hybrid algorithms (neural network-genetic algorithms and modified neuronal genetic algorithms). The results validate its effectiveness in both static and dynamic scenarios, offering a robust and intelligent approach to USV navigation in complex environments.