Robot Parkour Dataset

该数据集由安装在Unitree A1机器人上的Intel RealSense D435深度相机在不同跑酷任务中捕获的图像组成。该数据集支持在单个3090 GPU上训练，且训练时间不超过20小时，涵盖了包括高跳、远跳以及不同行走方式之间的转换等多种跑酷任务。此外，该数据集还适用于大规模的腿部机器人跑酷任务强化学习模拟训练，以及用于训练和评估神经网策略。

This dataset comprises images captured by an Intel RealSense D435 depth camera mounted on a Unitree A1 robot across diverse parkour tasks. It supports training on a single 3090 GPU with a total training time not exceeding 20 hours, and covers a variety of parkour tasks including high jumps, long jumps, and transitions between different walking gaits. Furthermore, this dataset is applicable to large-scale reinforcement learning simulation training for legged robot parkour tasks, as well as for training and evaluating neural network policies.