This article integrated reinforcement learning with fuzzy logic method for AUV local planning under the strong sea flow field. A fuzzy behavior is defined to resist the sea flow by giving a extra angle towards sea flow. And Q-learning is used to adjust the peak point of fuzzy membership function of the resisting sea flow behavior. This behavior is complemented by two other behaviors, the moving-to-goal behavior and collision avoiding behavior. The recommendations of these three behaviors are integrated through adjustable weighting factors to generate the final motion command for the AUV. Simulation shows it improve the adaptability of AUV under different sea flow greatly.
展开▼
