Research On Multi-USV Collision Avoidance Planning And Learning Method

When the Multi-USV performs autonomous operations in a complex and varied marine environment,whether it is a reef,an island,or some other dynamic object,it will pose a threat to the operation of the Multi-USV,and each USV in the cluster must maintain a certain degree during the navigation.Safety distance,the independent collision avoidance capability of the Multi-USV is one of the key technologies for its smooth and efficient completion of missions.This paper discusses the following aspects of the collaborative collision avoidance planning and autonomous learning methods of the Multi-USV: the working environment model of the Multi-USV,the internal collision avoidance planning method of the Multi-USV based on the improved Boid model and the speed adjustment strategy,and the improved particles based on the complex environment.Group-optimization algorithm Multi-USV cooperative collision avoidance and deep learning algorithm based Multi-USV autonomous collision avoidance learning.Firstly,based on the parameters of sensors such as existing navigation radar,photoelectric sensor and ship automatic identification system AIS,the USV is simulated to detect the surrounding environment,and the corresponding obstacle model,coordinate system and coordinates system conversion are constructed according to the characteristics of USV and its working environment.Secondly,the Boid model of the Multi-USV is constructed according to the three rules in the Boid model,and the three rules in the Boid model are converted into gravitation/repulsive methods.The Boid model is only a defect in the direction adjustment of the USV.Based on the Boid model,the speed adjustment coordination control strategy is introduced to better realize the coordination and collision avoidance between the Multi-USVs.The proposed coordination collision avoidance method is also verified by simulation.Then,the collision avoidance algorithm of Multi-USV in complex environment is designed.the integrated environmental information measured by the navigation radar and photoelectric sensors in real time is combined with the improved particle swarm optimization algorithm to plan the non-collision path in the USV integrated view in a rolling manner.It not only overcomes the shortcomings of the standard particle swarm optimization algorithm which is easy to fall into local optimum,but also combines the current environmental information of USV to improve the real-time performance of USV collision avoidance planning.In order to improve the smoothness of the path,USV is added to the fitness function.The corner optimization is carried out,and the above method is simulated and verified.Finally,the deep learning is briefly introduced,and the RevGRU-RNN network model is designed to realize the learning of the Multi-USV self-collision avoidance planning.In addition,the RevGRU-RNN network model is also derived,and the methods of processing sample data and improving network generalization are analyzed.The network model after training is simulated and verified.