This dataset is an Arabic-localized version of the standard lerobot/libero_10_image dataset. It is specifically designed to enable the training and fine-tuning of Vision-Language-Action (VLA) models (like OpenVLA or FastVLA) to perform complex multi-task robotic manipulation using native Arabic instructions.

Dataset Summary

Task: LIBERO-10 (Long-horizon multi-task manipulation) Localization: 100% Arabic… See the full description on the dataset page: https://huggingface.co/datasets/hamzabouajila/ar-libero-10-image.