Hugging Face Publishes ‘LIBERO’ Dataset for Robot Manipulation Training

The ‘physical-intelligence/libero’ dataset for imitation learning in robot manipulation has been published on Hugging Face’s dataset page. It includes camera images, states, actions, timestamps and other information.

The dataset is relevant to researchers who want to examine the relationship between actions and corresponding data using records that include not only images but also wrist images, states, actions and timestamps. The page lists the contents of each field, allowing researchers to see concretely what information they can use when working with training trajectories.