This dataset contains triplets of (current observation, action instruction, future observation) for training models to predict future frames of robotic actions.
Dataset Structure
Data Fields
current_frame: Input image (RGB) of the current observation
instruction: Textual description of the action to perform
future_frame: Target image (RGB) showing the expected outcome 50 frames later… See the full description on the dataset page: https://huggingface.co/datasets/bryandts/robotwin-action-prediction-dataset.