Research On Cooperative Danger Avoidance Method Of Multiple Unmanned Surface Vehicle

Unmanned Surface Vehicle(USV),a part of marine intelligent unmanned system,has been widely used in cooperative operation tasks such as maritime exploration,patrol and defense.The formation of unmanned system and danger avoidance technology are important for the successful implementation of maritime operation tasks.The complex and changeable environment of sea brings new challenges and requirements to the hazard avoidance technology of multi-unmanned vehicle formation.The cooperative hazard avoidance technology of multi-unmanned vehicle formation is studied in the thesis.As for the hazard avoidance technology of multiple USV formations,A distributed multiUSV reactive collaborative hazard avoidance method that conforms to the maritime collision avoidance rules is proposed in the thesis.This method mainly uses the distributed formation method to solve the formation coordination problem and regards the formation as a part.Rank the risk of obstacles by hazard.According to the dynamic window approach,maritime rules and other constraints,near-domain analysis and method of velocity obstacles are used to calculate the obstacle survival space.Then,get the optimal heading angle from the survival space.Afterwards,the optimal heading angles of different strategies is compared.After that,an appropriate method from to avoid danger from the strategies of maintaining formation,shrinking formation and changing formation is chosen.Subsequently,the effectiveness of the algorithm is verified by simulation experiments and sea trials.As for collision avoidance between formations of unmanned surface vehicle,considering that the USV cannot exchange information through real-time communication or under weak communication conditions.This paper research on the danger avoidance technology between USV based on perception information in complex scenarios.A method of danger avoidance among multiple USV based on deep reinforcement learning is proposed in the thesis.According to the real-time status obtained by the USV sensor and the information of near USV obtained by perception.State set of the risk avoidance algorithm is built.Considering safety reward anticollision distance function and conform to the USV movement set is designed.Reinforcement learning algorithm using depth training for USV collision between the value of the network.The multi-USV hazard avoidance experiment was carried out through simulation platform to verify the feasibility of the method.