海上布防任务中无人艇对可疑目标的驱逐方法研究

Expulsion of Suspicious Target by Unmanned Surface Vessels for Maritime Deployment Missions

海上无人船在海洋运输、环境调查、情报搜集等领域得到了日益广泛的应用。目前针对海上无人船的研究主要集中在目标跟踪、追逐避碰等问题,但布防任务中的驱逐任务有着更高要求,既要驱逐可疑船只远离保护目标,又要能够预测可疑船只运动意图,从而提前进行拦截,这给无人船的自主决策带来了挑战。针对上述问题,提出了对抗环境下海上无人船对可疑目标的驱逐方法。建立了基于深度强化学习的策略梯度优化算法框架;设计了意图预测模型及封堵策略,实现无人船的提前拦截,并证明了该策略的最优性;提出基于专家经验的矫正纠偏策略,降低了智能体早期的盲目探索时间,加快智能体的训练速度,并证明了专家动作序列的单值性;搭建了基于gym的仿真环境,并在仿真环境中验证了方法的有效性。研究结果表明,提出的驱逐方法能够使无人船在速度不占优势的情况下仍能完成对可疑目标的驱逐,为海上无人船在对抗性环境中执行复杂任务提供了一种技术参考。

Unmanned surface vessels(USVs)have been widely used in the fields of marine transportation,environment survey and information collection.The current research on USVs mainly focuses on target tracking and collision avoidance,and there are higher requirements for the eviction task in the deployment task.The USVs are required to not only expel suspicious targets away from a protection area,but also to predict the motion intention of the suspicious targets so as to intercept them in advance,which brings challenges to the autonomous decision making of USVs.To address this problem,we present an approach of deployment and expulsion of suspicious targets by USV in adversarial environments,and establish a DRL based framework for strategy gradient optimization.In order for the USV to intercept earlier by predicting the intention of suspicious targets,we design an intention prediction model and blocking strategy,and also prove the optimality of this strategy.A correction strategy based on expert experience is proposed to accelerate the training speed of the agents,and the single-valuedness of the expert action sequence has also been proved.A simulation environment is built based on gym,and the effectiveness of the method was verified in the simulation environment.The results show that the proposed expulsion method can enable the USV to complete the expulsion of suspicious targets without speed advantage,and it provides a technical reference for USVs to perform complex tasks in adversarial environments.