140 episodes · 31 fps · 4 cameras · 1280×720 av1
R1 Lite Specified Bread Bag — Masked
R1 Lite 양팔 로봇의 지정 빵봉투 집기(task: Pick up the specified bread bag)를 위한 GR00T 학습 데이터셋입니다.
원본 260908_pick_up_the_specified_bread_bag 데이터셋을 기반으로 하며, head camera 영상에 object-centric masking을 적용했습니다.
Task
여러 개의 빵봉투가 놓인 환경에서 instruction으로 지정된 위치의 빵봉투를 집습니다.
- Target position:
left,middle,right - Manipulation hand:
left,right - Total episodes: 140
Masking policy
각 head RGB 프레임에서 YOLO detector로 빵봉투 후보를 검출합니다.
- instruction의
left / middle / right에 따라 target bread bag을 선택합니다. - 선택되지 않은 빵봉투의 bounding box 영역을 검정색으로 마스킹합니다.
- target bread bag은 원본 이미지 그대로 유지합니다.
따라서 모델은 지정 대상에 집중한 head-view 입력을 사용해 행동을 예측합니다.
Modalities
| Modality | Processing |
|---|---|
| Head RGB | Target 외 빵봉투 bbox를 검정 마스킹 |
| Left wrist RGB | Original |
| Right wrist RGB | Original |
| Robot state / action | Original |
| Language instruction | Original |
Format
This dataset is converted to the GR00T training format and is intended for use with NEW_EMBODIMENT.
Intended use
- Object-centric visual conditioning for robot manipulation
- Comparison with the unmasked specified-bread-bag baseline
- Evaluation of target selection and distraction robustness in multi-object scenes
Notes
The same target-selection and YOLO masking pipeline must be applied to the live head-camera stream during robot inference.
This keeps the visual input distribution consistent between training and deployment.
- Downloads last month
- 180