NeRF Training Dataset

该数据集包含了大约1000帧，来自于一段5至6分钟的视频，用于训练人形机器人基于视觉导航和推球任务的政策。此外，该数据集还用于学习场景的接触几何，并利用神经辐射场（NeRF）技术进行新视角的合成。规模大约为1000帧，任务是将双足运动技能从模拟环境转移到真实环境。This dataset contains approximately 1000 frames extracted from a 5- to 6-minute video, and is primarily used for training policies for humanoid robots' visual navigation and ball-pushing tasks. Additionally, it is employed to learn the contact geometry of the scene and synthesize novel viewpoints using Neural Radiance Field (NeRF) technology. With around 1000 frames in total, this dataset also supports the transfer of bipedal locomotion skills from simulation environments to the real world.