spikefly/SemanticVLA-TraceX-240K-DROID

SemanticVLA TraceX 240K DROID数据集是TraceX-240K语料库的DROID组件，源自SemanticVLA项目，是一个带有轨迹注释的机器人轨迹数据集。该数据集以LeRobot v3.0格式重新打包了Open-X-Embodiment DROID v1.0.1版本，并包含密集的每帧末端执行器轨迹标签，直接存储在每个帧行中。数据集包含56,362个episodes和16,142,421帧，帧率为15 FPS，机器人类型为franka。它提供三个视频流：两个外部摄像头（exterior_image_1_left和exterior_image_2_left）和一个腕部摄像头（wrist_image_left）。轨迹标签包括trace.x和trace.y，表示归一化到[0, 100]尺度的图像空间坐标，其中(0,0)为左上角，(100,100)为右下角。数据用于机器人学习和视觉语言动作研究，覆盖了从DROID源数据集中选择的56K密集轨迹投影。The DROID component of TraceX-240K, a trace-annotated trajectory corpus introduced in SemanticVLA. This dataset is a LeRobot v3.0 repack of Open-X-Embodiment DROID v1.0.1 with dense per-frame end-effector trace labels stored directly in every frame row. It includes 56,362 episodes and 16,142,421 frames at 15 FPS, using a franka robot type. The dataset provides three video streams: two exterior cameras and one wrist camera. Trace labels consist of trace.x and trace.y, representing normalized image-space coordinates on a [0, 100] scale, with (0,0) as the top-left corner and (100,100) as the bottom-right. It is designed for robotics learning and visual-language-action research, covering 56K dense trace projections from the source DROID dataset.