OpenX-Sound OpenX-Sound is an audio-augmented robot manipulation dataset in RLDS / TFDS format. Each episode keeps the original trajectory structure and adds an episode-level audio field. Existing code that reads steps, observations, actions, language, and metadata can continue using the same paths, while audio-aware pipelines can additionally consume the synchronized waveform. Data Format The dataset is organized as standard TFDS directories: / /… See the full description on the dataset page:
Source: Hugging Face Hub (biubiu2/OpenX-Sound). Metadata imported from the dataset’s Hub tags.