Offline Target-Domain Trajectories for Cross-Domain AUV Dynamics Calibration and Policy Transfer

This dataset contains 50 task-independent target-domain trajectories collected from a physics-based three-dimensional, six-degree-of-freedom autonomous underwater vehicle hydrodynamics simulator. Each trajectory contains 300 control transitions and records the AUV pose, body velocity, actual actuator states, internal controller states, and normalized control commands. A rule-based multi-stage probing controller excites propulsion, horizontal turning, depth motion, and coupled actuator transients without using downstream task labels or rewards. Trajectory-specific ocean currents are sampled independently and remain fixed within each trajectory. Deterministic nested subsets containing 10, 30, and 50 trajectories are provided for evaluating offline dynamics-calibration methods under different data budgets. The package includes field definitions, units, collection and preprocessing protocols, target-domain parameters, random seeds, and experiment configurations.本数据集包含50条与任务无关的目标域航迹，采集自基于物理建模的三维六自由度自主水下航行器（Autonomous Underwater Vehicle，简称AUV）水动力学模拟器。每条航迹包含300个控制过渡步骤，并记录了AUV的位姿、本体速度、实际执行器状态、内部控制器状态以及归一化控制指令。本数据集使用基于规则的多阶段探测控制器，在不使用下游任务标签与奖励信号的前提下，激发推进、水平转向、深度运动以及耦合执行器瞬态响应。每条航迹对应的专属海流均独立采样，且在单条航迹内部保持恒定。数据集提供包含10、30与50条航迹的确定性嵌套子集，用于在不同数据规模预算下评估离线动力学校准方法。该数据集包包含字段定义、单位说明、数据采集与预处理规程、目标域参数、随机种子以及实验配置信息。