Mission planning for autonomous marine vehicles is nontrivial due to the dynamic and uncertain nature of the marine environment. Communication can be low‐bandwidth and is not always guaranteed, so the operator must rely on the vehicles to adjust their plans according to the realized state of the environment. This paper presents the improvements made to an energy‐aware mission planner that allows it to generate and adjust plans for an autonomous surface vehicle (ASV) operating in an uncertain environment. The energy‐aware mission planning problem was redefined as a stochastic programming problem, and a two‐stage solver was developed to provide an initial plan for the ASV and then adjust it during run‐time according to predefined recourse actions. The mission planner and ASV were trialed in Lake Waverley, Tasmania. Adjusting the recourse action criteria demonstrated that the ASV could exhibit conservative or opportunistic behaviors according to the operator's preference of safety margin. In the pursuit of extending the planner's second‐stage so that it can predict a suitable recourse action ahead of time, a hybrid long short‐term memory energy forecaster was trained from the Waverley mission data. Comparison of the error between the forecaster and the test data shows that the forecaster has a reliable forecast horizon of about 10 s.

中文翻译：

---

在无人驾驶水面车辆上实施的具有能源意识的任务计划器的现场试验

由于海洋环境的动态性和不确定性，用于自动船用车辆的任务计划并非易事。通信可以是低带宽的，并且不能总是得到保证，因此操作员必须依靠车辆根据已实现的环境状况调整计划。本文介绍了对能量感知型任务计划程序的改进，使其可以为在不确定环境中运行的自动水面车辆（ASV）生成和调整计划。能量敏感的任务计划问题被重新定义为随机规划问题，并且开发了两阶段求解器以为ASV提供初始计划，然后根据预定义的求助措施在运行时对其进行调整。任务计划者和ASV在塔斯马尼亚州的韦弗利湖试用。调整追索权行为标准表明，ASV可根据运营商的安全裕度偏好表现出保守或机会主义的行为。为了扩展计划者的第二阶段，以便可以提前预测合适的追索行动，从Waverley任务数据中训练了一个混合型长期短期记忆能量预测器。预测器与测试数据之间的误差比较表明，预测器具有约10 s的可靠预测范围。根据Waverley任务数据训练了一个混合型长期短期记忆能量预测器。预测器与测试数据之间的误差比较表明，预测器具有约10 s的可靠预测范围。根据Waverley任务数据训练了一个混合型长期短期记忆能量预测器。预测器与测试数据之间的误差比较表明，预测器具有约10 s的可靠预测范围。