内河无人驾驶船舶驾驶行为决策模型研究

Research on Navigation Behavior Decision Model of Unmanned Ship in Inland Waters

随着计算机技术的快速发展,基于机器学习算法的人工智能已经广泛运用于无人机和无人驾驶汽车领域研究.对于水上运输,由于船舶操纵性能相对特殊,无人驾驶船舶的研究起步较晚.文中采用强化学习算法,分别从状态空间、动作空间、回报函数和动作选择策略四个要素构建内河无人驾驶船舶驾驶行为决策模型,动作选择遵循《中华人民共和国内河避碰规则(2003修正版)》.通过与环境的不断交互学习积累经验,利用不断迭代的状态-动作对的回报值函数来实现驾驶行为决策的优化.分别对对驶相遇和交叉相遇两种会遇场景进行仿真.结果表明:该模型能较好地实现有人驾驶船舶一般船长的驾驶行为决策功能.

Using reinforcement learning algorithm,the navigation behavior decision-making model of unmanned ships in inland rivers was constructed from four elements:state space,action space,reward function and action selection strategy,and the action selection follows the Rules of the People's Republic of China for Preventing Collisions in Inland Rivers(Revised Edition 2003).Experience accumulation was realized through continuous interactive learning with environment,and optimization of navigation behavior decision was realized by using the return value function of iterative state-action pairs.Two encounter scenarios,head on situation and crossing situation,were simulated respectively.The results show that the model can well realize the decision-making function of the general captain's navigation behavior in manned ships.