real01b-square-d2-r3-fixedinit-checkpoint100000-heldout-sobol50

Real01b Square D2 R3 Heldout Sobol50 (TsFile) 是一个基于Apache TsFile格式的机器人学时间序列数据集，来源于LeRobot v3.0 Franka评估数据集，专门用于square_d2任务。该数据集包含150个训练集episodes，总计45,837帧，采样率为每秒15帧，以时间序列形式记录了多维数据，包括机器人状态、动作、任务几何参数、干预信号和奖励信号等，但排除了两个AV1相机流。数据集旨在评估三个不同策略（baseline_nocf_r3_checkpoint_100000、ours_nocf_r3_checkpoint_100000和ours_freecf_r3_checkpoint_100000）在50个独立Sobol序列初始状态下的性能表现，成功率范围在54%至58%之间。数据包含102个列，涵盖时间戳、episode索引、任务索引、帧索引、状态向量、动作向量、任务几何参数和奖励信号等详细信息。数据集采用Apache-2.0许可证，适用于机器人控制、策略评估和时间序列分析等领域的研究。Real01b Square D2 R3 Heldout Sobol50 (TsFile) is a robotics time series dataset in Apache TsFile format, derived from the LeRobot v3.0 Franka evaluation dataset for the square_d2 task. It contains 150 training episodes with a total of 45,837 frames, sampled at 15 fps, and records multi-dimensional data in time series form, including robot states, actions, task geometry, intervention signals, and reward signals, but excludes two AV1 camera streams. The dataset is used to evaluate the performance of three policies (baseline_nocf_r3_checkpoint_100000, ours_nocf_r3_checkpoint_100000, and ours_freecf_r3_checkpoint_100000) under 50 independent Sobol sequence initial states, with success rates ranging from 54% to 58%. The data includes 102 columns, covering timestamps, episode indices, task indices, frame indices, state vectors, action vectors, task geometry parameters, and reward signals. It is licensed under Apache-2.0 and suitable for research in robotics control, policy evaluation, and time series analysis.