基于速度矢量可行度的移动机器人多行为综合决策方法

  目的  为了实现无人驾驶船舶在未知环境下的智能避障功能，  方法  首先，建立一种基于深度强化学习（DRL）技术的无人驾驶船舶智能避碰决策模型，分析无人驾驶船舶智能避碰决策面临的问题，提出智能避碰决策的设计准则。然后，在此基础上，建立基于Markov决策方法（MDP）的智能避碰决策模型，通过值函数求解决策模型中的最优策略，使无人驾驶船舶状态对行为映射中的回报最大，并专门设计由接近目标、偏离航线和安全性组成的激励函数。
最后，分别在静态、动态障碍环境下进行仿真实验。  结果  结果表明，该智能决策方法可以有效避让障碍物，保障无人驾驶船舶在未知水域中的航行安全，  结论  所提方法可为无人驾驶船舶的自主航行提供理论参考。

Multi-behavior integrated-decision method based on feasibility of velocity vectors

  Objectives  In order to realize intelligent obstacle avoidance of unmanned vessel in unknown waters,  Methods  an intelligent obstacle avoidance decision-making model of the unmanned vessel based on Deep Reinforcement Learning(DRL)is established. Here we analyze the problems encountered in the unmanned vessel's intelligent obstacle avoidance decision-making, propose the design criteria of the intelligent obstacle avoidance decision-making, and then accordingly establish a decision-making model based on Markov Decision Process(MDP), through which obtain the optimal strategy by value function to make the maximum returns in behavior mapping of the unmanned vessel status and to design an excitation function specially composed of target approaching, off course and safety. Finally, carry out the simulation tests respectively in static and dynamic waters.
  Results  The results show that the proposed intelligent decision-making method can effectively avoid obstacles, and ensure the safe navigation of the unmanned vessel in unknown waters.  Conclusions  The proposed method can provide a theoretical reference for autonomous navigation of the unmanned vessel.