Research On Autonomy Improvement For Unmanned Surface Vehicle In Complicated Marine Environment

Unmanned surface vehicle (USV) is a kind of unmanned autonomous marine vehicle,with a certain degree of adaptive ability. With the development of unmanned systems,the intelligent level of USV is becoming higher and higher in recent years. But sometimes it is still unable to cope with some complicated environment and difficult tasks. Thus the cognitive capacity of human is needed to assist USV to cope with these difficulties. The adjustable autonomy mechanism in unmanned system makes USV could ask assistance from human by the way of human-robot interaction when meeting difficulties. The unmanned system of USV could extract the key information from user's intervention. Then the autonomy can be improved by learning this information.The main research work of this paper is to improve the autonomy for USV with the assistance from human. First of all, the division method of the autonomy level is introduced,and the way of human-robot interaction in each autonomy level is presented, which can be generally summarized as two kinds of interaction: 1) user provides the planning parameters to USV, and USV re-plan the task using these parameters; 2) user controls the speed and direction of USV in the form of manual operation. Based on these two kinds of interaction,this paper proposed two ways to improve the autonomy of USV.One method is the autonomy improvement for USV based on knowledge base. In this method,a framework of autonomy improvement is proposed,which has three modules:environmental monitoring, emergency recognition and knowledge base update. Then an emergency recognition algorithm is proposed, in which information gain is used to calculate the split points of emergency degree. According to the mission planning parameters provided by user and the uncertain event information, the knowledge base about navigation task is established. Then the knowledge base is updated according to certain rules. By using this knowledge base,USV could improve its autonomy.The other method is behavior imitation learning method based on inverse reinforcement learning. In this method,principal component analysis (PCA) is used to extract some behavior features from the expert strategy. Then the inverse reinforcement learning algorithm is used to calculate the weight of these features, and then the concrete expression of the reward function can also be calculated. Finally the policy iteration algorithm is used to obtain the optimal strategy that is similar to the expert strategy. Thus,the purpose of imitating expert behavior is achieved.Finally, the experiment is conducted to verify the above two methods of autonomy improvement in simulation experiment platform. In the experiment,USV executes the navigation task, and many kinds of uncertain factors are considered. It is can be seen clearly from experiment results and data analysis that, the two methods proposed in this paper could effectively improve the ability for USV to deal with the emergency, and improve the autonomy for USV.