BAIR Robot Pushing

该数据集包含了一系列Sawyer机器人手臂执行推举动作的视频，由于机器人动作具有随机性，因此存在多条合理的运动轨迹。此外，该数据集的无动作版本被用于训练VideoFlow模型，该模型能够基于过去帧的上下文生成目标帧。该数据集的分辨率为64x64，任务是对视频进行无监督生成。

This dataset contains a series of videos showing the Sawyer robotic arm performing lifting and pushing movements. Given the stochastic nature of the robot's movements, multiple plausible motion trajectories exist. Additionally, the action-free variant of this dataset was employed to train the VideoFlow model, which is capable of generating target frames using the context from preceding video frames. The dataset has a resolution of 64×64, with the task focusing on unsupervised video generation.

搜集汇总

背景与挑战

背景概述

BAIR Robot Pushing是伯克利人工智能研究院(BAIR)提供的机器人交互数据集，采用tfrecords格式存储，部分数据集包含数据结构的说明文件，遵循CC BY 4.0许可协议。