Information, computing and intelligent systems, Journal Year: 2024, Volume and Issue: 5, P. 4 - 16
Published: Dec. 26, 2024
Water drones have significant potential for use in environmental monitoring, search and rescue operations, marine infrastructure inspection, but the specific conditions of water environment make it difficult to implement stable autonomous navigation. The object research presented this paper is machine learning process navigation a drone model simulated environment. purpose study neural network using reinforcement method that provides improved obstacle avoidance adaptation currents. To achieve purpose, new based on proposed, which differs from existing ones uses an control algorithm takes into account speed direction current, makes possible stabilize generating coefficients. ensure effective optimization model, simulation training was developed USVSim simulator, contains various factors interfere with drone's movement, such as current presence other objects. drone, acting agent, gradually learns choose most actions maximize positive rewards through trial error, interacting adapting changing conditions. This place Deep Q-Network: value its state deep network; processes data, predicts action, gives agent. information form set sensor readings measuring distance nearest obstacles, drone’s heading goal. action received converted command rudder can understand. thruster power calculated by separate formulas trigonometric functions. results showed proposed allows decisions dynamic when rapid changes required. successfully adapted strategy feedback environment, so be concluded implemented shows further applications field drones, especially unpredictable environments.
Language: Английский
