kiroaiseoul/task04_pour_liquid_from_tubes_to_beaker_gist
task04_pour_liquid_from_tubes_to_beaker_gist GIST AI Lab이 수집한 11스테이지 실험실 프로토콜 데이터 중 stage 4 입니다. kiroaiseoul 네이밍(task04_pour_liquid_from_tubes_to_beaker)에 맞춰 재배포한 사본이고, 원본 배포명은 task4-pour-tubes-into-leftmost-beaker 입니다. Task: "Pour the solution from both tubes into the leftmost beaker" Episodes: 2,040 / Frames: 763,573 FPS: 30 · robot_type: mobileai_robot · codebase_version: v3.0 Cameras: cam_high, cam_left_wrist, cam_right_wrist (480x640x3, av1) action / observation.state: 16D… See the full description on the dataset page: https://huggingface.co/datasets/kiroaiseoul/task04_pour_liquid_from_tubes_to_beaker_gist.
task04pourliquidfromtubestobeaker_gist
GIST AI Lab이 수집한 11스테이지 실험실 프로토콜 데이터 중 stage 4 입니다. kiroaiseoul 네이밍(task04_pour_liquid_from_tubes_to_beaker)에 맞춰 재배포한 사본이고, 원본 배포명은 task4-pour-tubes-into-leftmost-beaker 입니다.
- Task: "Pour the solution from both tubes into the leftmost beaker"
- Episodes: 2,040 / Frames: 763,573
- FPS: 30 · robot_type: mobileairobot · **codebaseversion:** v3.0
- Cameras: camhigh, camleftwrist, camright_wrist (480x640x3, av1)
- action / observation.state: 16D (Mobile AI — base 2 + left 7 + right 7)
labels.csv
원본 배포에 동봉돼 있던 에피소드 단위 라벨이다. LeRobot 표준 파일이 아니라 repo 루트에 따로 둔다. 컬럼은 episode_id, task_scenario, success, data_issue.
data_issue 는 전 스테이지 통틀어 mobile_base_not_captured 한 종류뿐이다 — 모바일 베이스 상태가 고정값으로 기록돼 base 2D가 무의미해진 에피소드다. 팔 동작만 쓰는 경우엔 문제가 없지만 base를 학습에 넣는다면 걸러야 한다.
import pandas as pd
lab = pd.read_csv("labels.csv")
keep = lab[(lab.success == 1) & (lab.data_issue.isna())]
episodes = sorted(int(e.removeprefix("ep_")) for e in keep.episode_id)