robo_ai_u_r5e

  • Description:

Universal Robots UR5e demonstration dataset for VLA model training. Data collected from real-world robot workspace with preprogrammed robot routines.

Split Examples
  • Feature structure:
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'episode_index': int32,
        'file_name': string,
    }),
    'steps': Dataset({
        'absolute_action_mask': Tensor(shape=(7,), dtype=bool),
        'action': Tensor(shape=(7,), dtype=float32),
        'action_normalization_mask': Tensor(shape=(7,), dtype=bool),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'language_instruction': string,
        'observation': FeaturesDict({
            'image': Image(shape=(480, 640, 3), dtype=uint8),
            'robot_state': Tensor(shape=(15,), dtype=float32),
        }),
    }),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_metadata FeaturesDict
episode_metadata/episode_index Tensor int32
episode_metadata/file_name Tensor string
steps Dataset
steps/absolute_action_mask Tensor (7,) bool
steps/action Tensor (7,) float32
steps/action_normalization_mask Tensor (7,) bool
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/language_instruction Tensor string
steps/observation FeaturesDict
steps/observation/image Image (480, 640, 3) uint8
steps/observation/robot_state Tensor (15,) float32 
@article{RoboAIUR5e2025,
  title        = {RoboAI UR5e Training Dataset},
  author       = {Joonas Rouhiainen},
  institution  = {RoboAI Research Center, Satakunta University of Applied Sciences},
  year         = {2025},
  howpublished = {\url{https://sites.google.com/view/roboai-ur5e/home} },
}