Enhancing USV Navigation through Adaptive Obstacle Avoidance and Improved Q-Learning Path Planning

Abstract

To achieve adaptive navigation for Unmanned Surface Vehicles (USVs) in specific environments, a model for automatic obstacle avoidance path planning based on an enhanced Q-Learning algorithm has been developed. This model accounts for the unique maneuvering characteristics of USVs, intensive learning processes, and spatial environmental factors. Enhancements to the Q-Learning algorithm include modifications to the Q-value update method, optimization of the reward path configuration during the USV's avoidance path planning, and overall improvements in algorithm learning efficiency. The action selection is tailored to the operational characteristics of USVs, and path optimization is conducted based on the USV's operational cycle. A simulation environment was created using the MATLAB GUI platform to test the model. Results from these simulations indicate that the USV, employing this intensive learning approach, can effectively plan superior obstacle avoidance routes that align with the vehicle's motion characteristics and successfully navigate around multiple obstacles.