vima_converted_externally_to_rlds

  • Keterangan :

Kumpulan data SIM dari satu lengan robot yang melakukan tugas meja yang dihasilkan secara prosedural dengan perintah multimodal, 600 ribu+ lintasan

Membelah Contoh
  • Struktur fitur :
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'action_bounds': FeaturesDict({
            'high': Tensor(shape=(3,), dtype=float32),
            'low': Tensor(shape=(3,), dtype=float32),
        }),
        'end-effector type': string,
        'failure': Scalar(shape=(), dtype=bool),
        'file_path': string,
        'n_objects': Scalar(shape=(), dtype=int64),
        'num_steps': Scalar(shape=(), dtype=int64),
        'robot_components_seg_ids': Sequence(Scalar(shape=(), dtype=int64)),
        'seed': Scalar(shape=(), dtype=int64),
        'success': Scalar(shape=(), dtype=bool),
        'task': string,
    }),
    'steps': Dataset({
        'action': FeaturesDict({
            'pose0_position': Tensor(shape=(3,), dtype=float32),
            'pose0_rotation': Tensor(shape=(4,), dtype=float32),
            'pose1_position': Tensor(shape=(3,), dtype=float32),
            'pose1_rotation': Tensor(shape=(4,), dtype=float32),
        }),
        'discount': Scalar(shape=(), dtype=float32),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'multimodal_instruction': string,
        'multimodal_instruction_assets': FeaturesDict({
            'asset_type': Sequence(string),
            'frontal_image': Sequence(Tensor(shape=(128, 256, 3), dtype=uint8)),
            'frontal_segmentation': Sequence(Tensor(shape=(128, 256), dtype=uint8)),
            'image': Sequence(Tensor(shape=(128, 256, 3), dtype=uint8)),
            'key_name': Sequence(string),
            'segmentation': Sequence(Tensor(shape=(128, 256), dtype=uint8)),
            'segmentation_obj_info': Sequence({
                'obj_name': Sequence(string),
                'segm_id': Sequence(Scalar(shape=(), dtype=int64)),
                'texture_name': Sequence(string),
            }),
        }),
        'observation': FeaturesDict({
            'ee': int64,
            'frontal_image': Tensor(shape=(128, 256, 3), dtype=uint8),
            'frontal_segmentation': Tensor(shape=(128, 256), dtype=uint8),
            'image': Tensor(shape=(128, 256, 3), dtype=uint8),
            'segmentation': Tensor(shape=(128, 256), dtype=uint8),
            'segmentation_obj_info': FeaturesDict({
                'obj_name': Sequence(string),
                'segm_id': Sequence(Scalar(shape=(), dtype=int64)),
                'texture_name': Sequence(string),
            }),
        }),
        'reward': Scalar(shape=(), dtype=float32),
    }),
})
  • Dokumentasi fitur :
Fitur Kelas Membentuk Tipe D Keterangan
FiturDict
episode_metadata FiturDict
episode_metadata/action_bounds FiturDict
episode_metadata/action_bounds/high Tensor (3,) float32
episode_metadata/action_bounds/rendah Tensor (3,) float32
tipe episode_metadata/efektor akhir Tensor rangkaian
episode_metadata/kegagalan Skalar bodoh
episode_metadata/file_path Tensor rangkaian
episode_metadata/n_objects Skalar int64
episode_metadata/num_steps Skalar int64
episode_metadata/robot_components_seg_ids Urutan (Skalar) (Tidak ada,) int64
episode_metadata/seed Skalar int64
episode_metadata/sukses Skalar bodoh
episode_metadata/tugas Tensor rangkaian
Langkah Himpunan data
langkah/tindakan FiturDict
langkah/tindakan/pose0_position Tensor (3,) float32
langkah/tindakan/pose0_rotation Tensor (4,) float32
langkah/tindakan/pose1_posisi Tensor (3,) float32
langkah/tindakan/pose1_rotasi Tensor (4,) float32
langkah/diskon Skalar float32
langkah/adalah_pertama Tensor bodoh
langkah/adalah_terakhir Tensor bodoh
langkah/is_terminal Tensor bodoh
langkah/multimodal_instruction Tensor rangkaian
langkah/multimodal_instruction_assets FiturDict
langkah/multimodal_instruction_assets/asset_type Urutan (Tensor) (Tidak ada,) rangkaian
langkah/multimodal_instruction_assets/frontal_image Urutan (Tensor) (Tidak ada, 128, 256, 3) uint8
langkah/multimodal_instruction_assets/frontal_segmentation Urutan (Tensor) (Tidak ada, 128, 256) uint8
langkah/multimodal_instruction_assets/image Urutan (Tensor) (Tidak ada, 128, 256, 3) uint8
langkah/multimodal_instruction_assets/key_name Urutan (Tensor) (Tidak ada,) rangkaian
langkah/multimodal_instruction_assets/segmentasi Urutan (Tensor) (Tidak ada, 128, 256) uint8
langkah/multimodal_instruction_assets/segmentation_obj_info Urutan
langkah/multimodal_instruction_assets/segmentation_obj_info/obj_name Urutan (Tensor) (Tidak ada,) rangkaian
langkah/multimodal_instruction_assets/segmentation_obj_info/segm_id Urutan (Skalar) (Tidak ada,) int64
langkah/multimodal_instruction_assets/segmentation_obj_info/texture_name Urutan (Tensor) (Tidak ada,) rangkaian
langkah/pengamatan FiturDict
langkah/pengamatan/ee Tensor int64
langkah/pengamatan/frontal_image Tensor (128, 256, 3) uint8
langkah/pengamatan/frontal_segmentasi Tensor (128, 256) uint8
langkah/pengamatan/gambar Tensor (128, 256, 3) uint8
langkah/observasi/segmentasi Tensor (128, 256) uint8
langkah/pengamatan/segmentasi_obj_info FiturDict
langkah/pengamatan/segmentasi_obj_info/obj_name Urutan (Tensor) (Tidak ada,) rangkaian
langkah/pengamatan/segmentasi_obj_info/segm_id Urutan (Skalar) (Tidak ada,) int64
langkah/pengamatan/segmentasi_obj_info/texture_name Urutan (Tensor) (Tidak ada,) rangkaian
langkah/hadiah Skalar float32
@inproceedings{jiang2023vima,  title     = {VIMA: General Robot Manipulation with Multimodal Prompts},  author    = {Yunfan Jiang and Agrim Gupta and Zichen Zhang and Guanzhi Wang and Yongqiang Dou and Yanjun Chen and Li Fei-Fei and Anima Anandkumar and Yuke Zhu and Linxi Fan}, booktitle = {Fortieth International Conference on Machine Learning},  year      = {2023}. }