vima_converted_externally_to_rlds

  • 설명 :

다중 모드 프롬프트, 600K+ 궤적을 사용하여 절차적으로 생성된 테이블 작업을 수행하는 단일 로봇 팔의 SIM 데이터세트

나뉘다
  • 기능 구조 :
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),
    }),
})
  • 기능 문서 :
특징 수업 모양 Dtype 설명
특징Dict
에피소드_메타데이터 특징Dict
Episode_metadata/action_bounds 특징Dict
Episode_metadata/action_bounds/high 텐서 (삼,) float32
Episode_metadata/action_bounds/low 텐서 (삼,) float32
Episode_metadata/엔드 이펙터 유형 텐서
Episode_metadata/실패 스칼라 부울
에피소드_메타데이터/파일_경로 텐서
Episode_metadata/n_objects 스칼라 정수64
Episode_metadata/num_steps 스칼라 정수64
Episode_metadata/robot_comComponents_seg_ids 시퀀스(스칼라) (없음,) 정수64
에피소드_메타데이터/시드 스칼라 정수64
에피소드_메타데이터/성공 스칼라 부울
에피소드_메타데이터/작업 텐서
단계 데이터세트
단계/작업 특징Dict
단계/액션/pose0_position 텐서 (삼,) float32
단계/액션/pose0_rotation 텐서 (4,) float32
단계/액션/pose1_position 텐서 (삼,) float32
단계/액션/pose1_rotation 텐서 (4,) float32
걸음수/할인 스칼라 float32
걸음수/is_first 텐서 부울
걸음수/is_last 텐서 부울
단계/is_terminal 텐서 부울
단계/multimodal_instruction 텐서
단계/multimodal_instruction_assets 특징Dict
단계/multimodal_instruction_assets/asset_type 시퀀스(텐서) (없음,)
단계/multimodal_instruction_assets/frontal_image 시퀀스(텐서) (없음, 128, 256, 3) uint8
단계/multimodal_instruction_assets/frontal_segmentation 시퀀스(텐서) (없음, 128, 256) uint8
단계/multimodal_instruction_assets/이미지 시퀀스(텐서) (없음, 128, 256, 3) uint8
단계/multimodal_instruction_assets/key_name 시퀀스(텐서) (없음,)
단계/multimodal_instruction_assets/분할 시퀀스(텐서) (없음, 128, 256) uint8
단계/multimodal_instruction_assets/segmentation_obj_info 순서
단계/multimodal_instruction_assets/segmentation_obj_info/obj_name 시퀀스(텐서) (없음,)
단계/multimodal_instruction_assets/segmentation_obj_info/segm_id 시퀀스(스칼라) (없음,) 정수64
단계/multimodal_instruction_assets/segmentation_obj_info/texture_name 시퀀스(텐서) (없음,)
단계/관찰 특징Dict
단계/관찰/ee 텐서 정수64
단계/관찰/frontal_image 텐서 (128, 256, 3) uint8
단계/관찰/frontal_segmentation 텐서 (128, 256) uint8
단계/관찰/이미지 텐서 (128, 256, 3) uint8
단계/관찰/분할 텐서 (128, 256) uint8
단계/관찰/segmentation_obj_info 특징Dict
단계/관찰/segmentation_obj_info/obj_name 시퀀스(텐서) (없음,)
단계/관찰/segmentation_obj_info/segm_id 시퀀스(스칼라) (없음,) 정수64
단계/관찰/segmentation_obj_info/texture_name 시퀀스(텐서) (없음,)
걸음수/보상 스칼라 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}. }