OpenX-Sound
收藏资源简介:
OpenX-Sound是一个音频增强的机器人操作数据集,采用RLDS(机器人学习数据集标准)/TFDS(TensorFlow数据集)格式。它在标准机器人操作轨迹数据的基础上,为每个轨迹新增了一个同步的完整波形音频字段(`audio`),同时完整保留了原始的`steps`数据结构(包含观察、动作、语言指令、元数据等)。这种设计使得现有仅处理传统机器人轨迹的代码无需修改即可继续运行,而支持多模态处理的流程则可以额外利用同步的音频信号。数据集旨在服务于音频感知的机器人学习、多模态轨迹建模以及声音条件操作等研究方向,支持研究者将音频信号与视觉观察、语言指令和机器人动作进行联合分析与建模。数据集文件以标准TFDS目录结构组织发布,并遵循CC-BY-NC-4.0许可协议。
OpenX-Sound is an audio-enhanced robot manipulation dataset, formatted in RLDS (Robotics Learning Dataset Standard)/TFDS (TensorFlow Dataset) format. Based on standard robot manipulation trajectory data, it adds a synchronized full-waveform audio field (`audio`) for each trajectory (episode), while fully preserving the original `steps` data structure (including observations, actions, language instructions, metadata, etc.). This design allows existing code that only processes traditional robot trajectories to continue running without modification, while multimodal processing pipelines can additionally leverage the synchronized audio signals. The dataset aims to serve research directions such as audio-aware robot learning, multimodal trajectory modeling, and sound-conditioned manipulation, supporting researchers in jointly analyzing and modeling audio signals with visual observations, language instructions, and robot actions. The dataset files are organized and released in the standard TFDS directory structure and follow the CC-BY-NC-4.0 license agreement.




